A Comparative Analysis of Machine Learning Classification Techniques in the Prediction of Autism in Children

A Comparative Analysis of Machine Learning Classification Techniques in the Prediction of Autism in Children | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Comparative Analysis of Machine Learning Classification Techniques in the Prediction of Autism in Children Akintayo Ayoade, Ebierimunu Abule, Chisom Onwugbenu, Idowu Olugbenga Adewumi This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7779588/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Autism Spectrum Disorder (ASD) impacts around 1 in 100 children worldwide, but prompt diagnosis is still limited due to subjective clinical assessments. This research compared nine supervised machine learning (ML) classifiers: Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and ensemble methods such as Bagging, Boosting (XGBoost), and Stacking to predict ASD in children. Two artificially created datasets were used: one imbalanced dataset with 50,000 samples having 10% positive cases (5,000 autistic, 45,000 non-autistic), and one balanced dataset of the same size with 50% positive cases (25,000 autistic, 25,000 non-autistic). Every dataset included 19 features covering demographics (3 attributes), parental/medical history (3 attributes), behavioral screening items (10 binary responses), one combined score, and a binary target label. Metrics for evaluation comprised accuracy, precision, recall, F1-score, AUROC, and AUPRC. In the imbalanced dataset, RF reached an F1-score of 0.75, AUROC of 0.91, and AUPRC of 0.66, surpassing LR (F1 = 0.51) and KNN (F1 = 0.53). SVM and MLP closely trailed with F1-scores ranging from 0.71 to 0.73. In the balanced dataset, ensemble models notably enhanced performance: Stacking attained an F1-score of 0.91, AUROC of 0.96, and AUPRC of 0.95, whereas Boosting yielded F1 = 0.90 and AUROC = 0.95. Baseline models like LR and DT showed moderate improvements, achieving F1-scores of approximately 0.80–0.81. Statistical validation through McNemar’s test revealed significant differences (p = 0.040) between RF and SVM in imbalanced circumstances. Analysis of computational efficiency showed differences in runtime, with LR finishing in 5.5 seconds, RF in 17.0 seconds, and MLP in 37.3 seconds. The findings indicated that ensemble models, especially Stacking and Boosting, deliver enhanced predictive accuracy and reliability across various class distributions, suggesting their possible incorporation into clinical decision support systems for scalable, data-informed early detection of ASD. Artificial Intelligence and Machine Learning Health Economics and Outcomes Research Autism Spectrum Disorder (ASD) Machine Learning Classification Ensemble Models Data Imbalance Early Detection Clinical Decision Support Systems (CDSS) Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Autism Spectrum Disorder (ASD) is a multifaceted neurodevelopmental condition marked by ongoing challenges in social communication, limited interests, and repetitive actions 1 . Timely interventions that can greatly enhance developmental outcomes depend on the early identification of ASD, but the diagnosis still relies heavily on subjective clinical evaluations 2 . Conventional diagnostic methods, including the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview-Revised (ADI-R), while dependable, demand considerable time, necessitate specialized skills, and are frequently unavailable in low-resource environments. As a result, there is an increasing demand for data-based screening instruments that can help healthcare providers and decision-makers in recognizing children at risk of ASD more effectively and impartially 3 . Recent developments in artificial intelligence (AI) and machine learning (ML) have shown encouraging abilities in recognizing intricate, nonlinear patterns from behavioral and demographic information 4 . Through the use of computational models, scientists have managed to create automated classifiers that anticipate ASD status based on characteristics obtained from questionnaires, observational checklists, and clinical documentation. These models not only improve the predictive precision of screening methods but also facilitate scalable approaches for early identification in various populations. Numerous studies have investigated the use of algorithms like Decision Trees (DT), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Networks (ANN) to categorize autism-related data, producing different levels of success based on dataset features and preprocessing methods 5 . Although advancements have been made, inconsistencies remain in the ideal model selection and feature representation for predicting ASD. Numerous earlier studies have depended on small datasets, inadequate feature engineering, or non-uniform evaluation methods, leading to models that excel on training data but fail to generalize effectively to new instances. Additionally, certain research has neglected factors like data imbalance, redundancy in features, and the thoroughness of cross-validation, which may lead to an exaggeration of performance metrics. Consequently, a thorough and methodologically rigorous comparison of various machine learning classifiers assessed under uniform preprocessing and validation conditions is crucial for setting dependable benchmarks for ASD prediction 6 . This research aims to address that deficiency by methodically assessing the effectiveness of various supervised learning algorithms, such as Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Multilayer Perceptron (MLP), along with ensemble techniques like Bagging, Boosting, and Stacking. Employing a publicly accessible dataset of children's behavioral and demographic information, the research investigates the predictive capabilities of each model using metrics including accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). The study also examines the impact of hyperparameter optimization methods (GridSearchCV and RandomizedSearchCV) and evaluates computational efficiency to determine the balance between model performance and execution time. This work's key contribution is offering a transparent and reproducible comparison of classification methods for predicting ASD within standardized experimental conditions. By pinpointing the most successful models and their key attributes, the results provide valuable guidance for creating smart, economical screening systems that can enhance conventional clinical evaluations. The results of this research are anticipated to aid in the development of early detection systems and data-informed strategies that enhance tailored care and policy formulation for children with autism. 2. Related Work 2.1 Overview of Existing Machine Learning Applications in ASD Detection The rise of machine learning (ML) has changed the field of autism spectrum disorder (ASD) research by allowing for automated and objective forecasting of diagnostic results from intricate behavioral and demographic data. Initial research showed that ML algorithms could effectively classify ASD cases utilizing standardized screening tools like the Autism Diagnostic Observation Schedule (ADOS) 7 , Autism Diagnostic Interview–Revised (ADI-R), and behavioral questionnaires reported by parents 8 . Utilizing these data sources, researchers have managed to uncover fundamental patterns and relationships that are frequently hard to detect using traditional statistical techniques. These models have the capability to improve screening precision, decrease diagnostic delays, and broaden access to early detection resources, especially in low-resource healthcare environments 9 . Recent studies have highlighted the incorporation of computational intelligence methods like Decision Trees (DT), Support Vector Machines (SVM), Random Forests (RF), Artificial Neural Networks (ANN), and Logistic Regression (LR) for the identification of ASD 11 , 12 , 13 . Sarker et al. (2022) showed that ensemble classifiers surpass conventional techniques in differentiating neurotypical from ASD children when given meticulously preprocessed behavioral characteristics 14 . In the same vein, Thabtah (2017) created a mobile autism screening application employing rule-based classification, offering a concrete instance of how ML can enhance traditional evaluation methods. These advancements highlight the potential of data-informed methods in improving ASD screening practices among various demographic populations 15 . 2.2 Comparative Examination of Models Utilized in Previous Research An increasing amount of comparative studies has aimed to assess the effectiveness of ML algorithms in forecasting ASD 16 . Decision Tree and Random Forest models are commonly preferred due to their clarity and resistance to various data types. They excel with small to medium datasets and can identify non-linear connections between behavioral characteristics and diagnostic classifications. SVMs are commonly used because of their ability to manage high-dimensional feature spaces and reach large classification margins, although they often need significant parameter adjustment and are less clear for clinical analysis. Artificial Neural Networks (ANNs) and their variations, such as Multilayer Perceptrons (MLPs), exhibit exceptional learning abilities in intricate, nonlinear datasets; nonetheless, they require more extensive datasets and computational power. Naïve Bayes (NB), despite its simplicity, has been employed as a standard in numerous ASD studies to compare the effectiveness of advanced algorithms 17 . Various studies indicate differing levels of success among these models. For instance, Abbas et al. (2020) found that RF performed better than SVM and ANN in predicting ASD characteristics in children using datasets from questionnaires 18 . In contrast, Iqbal et al. (2021) discovered that when enhanced with sophisticated feature selection and regularization, MLP surpassed tree-based models 19 . Regardless of these outcomes, comparative results remain variable because of differences in dataset sizes, preprocessing methods, and evaluation protocols among various studies. 2.3 Overview of Datasets Utilized in Related Studies Many machine learning research efforts focused on predicting autism depend on publicly accessible datasets like those found on the UCI Machine Learning Repository, especially the “Autism Screening for Children” dataset created by Thabtah (2018) 20 . This dataset consists of behavioral and demographic factors gathered via standardized questionnaires and has acted as a standard for various classification research. Other studies have utilized clinically validated tools like the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview–Revised (ADI-R) to capture more detailed clinical characteristics. Nonetheless, the availability of these clinical datasets is frequently restricted by privacy and ethical considerations, hindering their application in extensive ML experiments. As a result, researchers often rely on UCI-derived or artificial datasets for reproducible and publicly accessible experimentation. Although these datasets have enhanced empirical research, they differ greatly in feature composition and representativeness. For example, datasets based on questionnaires focus on parental insights and socio-demographic details, while clinical datasets record systematic behavioral coding. This variability makes it difficult to compare studies and may affect the applicability of models to practical clinical situations 21 . 2.4 Recognized Issues in Previous Research Despite encouraging findings, current research highlights various issues that limit the effectiveness and real-world applicability of ML-driven ASD prediction. Data imbalance continues to be a significant challenge, as ASD instances are generally less represented compared to non-ASD samples, resulting in biased classifiers that prioritize the majority class. Methods like oversampling, Synthetic Minority Oversampling Technique (SMOTE), and cost-sensitive learning have been suggested, yet their implementation is still uneven. Another significant issue is feature leakage, where specific questionnaire items or created features directly capture the target label, thus artificially boosting performance metrics. Moreover, numerous studies are hindered by small sample sizes, which restrict the statistical power of the models and heighten the likelihood of overfitting. The absence of uniform preprocessing pipelines makes reproducibility even more challenging. Variations in feature encoding, scaling methods, and cross-validation techniques can lead to inconsistent results despite utilizing the identical dataset. Additionally, several studies depend on a single random train–test split instead of rigorous k-fold cross-validation, weakening the trustworthiness of reported metrics. 2.5 Gaps in Reproducibility, Verification, and Model Comprehensibility A significant drawback in current studies is the lack of reproducibility and external validation. Although reported accuracies frequently surpass 95%, a limited number of studies provide their code, random seeds, or detailed preprocessing scripts, complicating independent validation. Likewise, the lack of external validation through independent datasets restricts the applicability of these models to varied populations. A rising issue is the interpretability of models. Deep learning and ensemble methods, though effective, frequently function as "black boxes," providing minimal understanding of the behavioral or demographic factors influencing predictions. New frameworks like SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations) have emerged to improve interpretability, yet their application in ASD research is still limited. 2.6 Justification for Choosing Classification Algorithms in This Research In light of these recognized challenges, this research employs a comparative multi-model framework to assess a range of supervised learning algorithms following a uniform experimental protocol. The selected models, Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Multilayer Perceptron (MLP) illustrated a well-rounded range of linear, nonlinear, and ensemble techniques. This choice allows for an extensive evaluation of algorithmic performance across various data formats and complexity degrees. Ensemble techniques like Bagging, Boosting, and Stacking are additionally included to assess their potential for decreasing variance and improving accuracy. This research seeks to create clear and reproducible benchmarks for ASD prediction by systematically evaluating these algorithms within a standardized preprocessing pipeline and a thorough cross-validation framework. Additionally, by providing both performance indicators and computational efficiency, the research offers practical perspectives on model selection trade-offs, aiding the creation of efficient, understandable, and deployable screening systems for early detection of autism. 3. Materials and Methods 3.1 Dataset Description The experimental examination utilized a synthetically created dataset that replicated the Autism Screening for Children dataset from the University of California Irvine (UCI) Machine Learning Repository. Two versions of the dataset were generated to enable a comparative assessment of model performance across varying class distributions (Table 1 ). The initial variant, a skewed dataset, contained 50,000 instances with around 10% categorized as autistic (positive class). The second version, a balanced dataset, included 50,000 examples with an equal number of autistic and non-autistic labels (50% positive class). Every dataset included 19 attributes organized into specific categories (Table 2 ): (i) demographic attributes (age, gender, ethnicity), (ii) medical and parental background attributes (jaundice at birth, family history of autism, prior use of screening instruments), and (iii) behavioral screening answers (ten binary indicators, A1–A10, taken from autism questionnaire questions). An aggregated screening score feature, indicating the total of affirmative responses throughout A1–A10, was also incorporated. The aim variable was a binary label representing the probability of autism (1 = autistic, 0 = non-autistic). Since the dataset was artificially created for research, it lacks identifiable personal details, thus removing direct ethical concerns. However, ethical factors were considered when designing the data generation process to ensure it accurately represented real-world populations and clinical situations. The dataset is shared under an open research license and is meant exclusively for educational and benchmarking use. 3.2 Data Preprocessing Before training the model, the datasets underwent several preprocessing procedures. Given that the data generation method resulted in complete records, there were no missing values, and thus, no imputation was necessary. All categorical features (gender, race, jaundice, family background, and previous use of screening tools) were converted into numerical format using one-hot encoding to enable their application in machine learning models (Table 3 ). Behavioral characteristics (A1–A10) were naturally binary and did not need any transformation. Continuous variables like age and the combined screening score were standardized through z-score normalization to minimize bias for algorithms that are sensitive to feature scales, like Support Vector Machines and Neural Networks. To avoid label leakage, no additional features directly linked to the target were included apart from the screening score obtained from the questionnaire. Ultimately, the datasets were divided into training, validation, and test sets employing a stratified sampling technique to maintain class distributions. The dataset was divided with 70% designated for training, and 15% assigned to both validation and testing. 3.3 Algorithms for Machine Learning A variety of machine learning classifiers was utilized to guarantee a thorough comparative analysis. Logistic Regression (LR) was chosen as a baseline linear model, whereas K-Nearest Neighbors (KNN) and Decision Trees (DT) were added for their non-parametric characteristics. Random Forests (RF) and Support Vector Machines (SVM) were utilized as strong ensemble and margin-oriented learners, respectively. A Multi-Layer Perceptron (MLP) was utilized as a neural network model to capture nonlinear relationships (Table 4 ). Furthermore, methods of ensemble learning were investigated. Bagging (Bootstrap Aggregating) was used to diminish variance in high-variance models like DTs, while Boosting techniques (e.g., AdaBoost, Gradient Boosting) were employed to enhance performance through sequential error correction. A Stacking ensemble was employed to combine predictions from various base learners into a meta-classifier for enhanced generalization. Hyperparameter optimization was carried out utilizing both GridSearchCV (comprehensive search over defined parameter grids) and RandomizedSearchCV (random parameter sampling) with five-fold cross-validation to enhance model performance. 3.4 Assessment Metrics The evaluation of model performance utilized various classification metrics (Table 6 ). The Accuracy score was utilized as an overall performance measure, while Precision, Recall, and the F1-score were presented to more effectively reflect trade-offs in classification, especially in imbalanced datasets (Table 5 ). The Area Under the Receiver Operating Characteristic curve (AUROC) assessed the models' ability to discriminate across thresholds, while the Area Under the Precision–Recall Curve (AUPRC) was highlighted due to its clinical importance in identifying minority positive cases (autism) within imbalanced datasets (Table 6 ). Choosing these metrics guarantees a well-rounded evaluation of model performance in both balanced and imbalanced situations, facilitating a strong comparison of algorithm efficiency. 3.5 Setup for Experimentation All experiments took place in a Python 3.10 environment utilizing the Scikit-learn library for implementing machine learning models, along with NumPy and Pandas for preprocessing data. Ensemble techniques and hyperparameter tuning were also performed using the Scikit-learn framework. The computing setup included a 16-core processor, 32 GB of RAM, and lacked any dedicated GPU support (Table 7 ). A constant random seed (42) was used throughout all experiments to guarantee the consistency of results. Model training and evaluation were performed under the same experimental conditions to establish a fair foundation for comparative analysis. Table 1 Dataset Characteristics Dataset Variant Rows Features Positive Cases Negative Cases Positive Class (%) Imbalanced Autism Dataset 50,000 19 5,000 45,000 10% Balanced Autism Dataset 50,000 19 25,000 25,000 50% Table 2 Feature Categories and Descriptions Feature Category Features Description Demographic Age, Sex, Ethnicity Age (2–18 years), biological sex (male/female), ethnicity categories Parental/Medical Jaundice, Family History, App Use Birth jaundice, family history of autism, prior use of screening application Behavioral (Binary) A1–A10 Ten autism screening questionnaire items (0 = No, 1 = Yes) Aggregated Behavior Screening Score Sum of positive responses across A1–A10 (range: 0–10) Target Label Label Autism likelihood (0 = No Autism, 1 = Autism) Table 3 Data Preprocessing Steps Step Approach Applied Missing Values None present (synthetically generated dataset) Categorical Encoding One-hot encoding for categorical variables (sex, ethnicity, jaundice, etc.) Feature Scaling Z-score normalization applied to age and screening score Feature Redundancy Removal No direct label-leaking features included beyond derived score Data Partitioning Stratified split: 70% training, 15% validation, 15% testing Table 4 Machine Learning Algorithms and Key Hyperparameters Algorithm Key Hyperparameters Tuned Logistic Regression (LR) Regularization type (L1/L2), C (inverse regularization strength) K-Nearest Neighbors (KNN) Number of neighbors (k), distance metric (Euclidean, Manhattan) Decision Tree (DT) Max depth, min samples per leaf, criterion (Gini, Entropy) Random Forest (RF) Number of trees, max depth, max features, bootstrap sampling Support Vector Machine (SVM) Kernel type (linear, RBF), C, gamma Multi-Layer Perceptron (MLP) Number of hidden layers, neurons per layer, activation function, learning rate Table 5 Ensemble Methods Employed Ensemble Method Base Learners Used Description Bagging Decision Trees, Random Forests Reduces variance by averaging predictions from bootstrap samples Boosting Decision Trees (shallow) Sequentially corrects errors by re-weighting misclassified samples Stacking Logistic Regression (meta-model) Combines predictions from multiple classifiers into a higher-level learner Table 6 Evaluation Metrics Metric Formula / Definition Purpose Accuracy (TP + TN) / (TP + FP + TN + FN) Overall correctness of predictions Precision TP / (TP + FP) Fraction of predicted positives that are correct Recall TP / (TP + FN) Fraction of actual positives correctly identified F1-Score 2 × (Precision × Recall) / (Precision + Recall) Harmonic mean of precision and recall AUROC Area under ROC curve (TPR vs. FPR across thresholds) Discriminative ability of classifier AUPRC Area under Precision–Recall curve Better assessment under class imbalance Table 7 Experimental Setup Component Specification Programming Language Python 3.10 Libraries Used Scikit-learn, NumPy, Pandas Computational Resources 16-core CPU, 32 GB RAM, no GPU acceleration Cross-Validation 5-fold stratified cross-validation Hyperparameter Search GridSearchCV, RandomizedSearchCV Reproducibility Fixed random seed (42) applied across all experiments 4. Results These section will be for displaying outcomes of the modeling and analysis carried out to achieve the aim and objectives of this research work. Table 8 Performance of Baseline Classifiers on Imbalanced Dataset (10% Positive Class) Classifier Accuracy Precision Recall F1-score AUROC AUPRC Logistic Regression (LR) 0.88 0.64 0.42 0.51 0.81 0.47 K-Nearest Neighbors (KNN) 0.85 0.60 0.48 0.53 0.78 0.44 Decision Tree (DT) 0.86 0.62 0.55 0.58 0.80 0.46 Random Forest (RF) 0.93 0.79 0.71 0.75 0.91 0.66 Support Vector Machine (SVM) 0.91 0.75 0.67 0.71 0.89 0.63 Multi-Layer Perceptron (MLP) 0.92 0.77 0.69 0.73 0.90 0.65 Table 9 Performance of Baseline Classifiers on Balanced Dataset (50% Positive Class) Classifier Accuracy Precision Recall F1-score AUROC AUPRC Logistic Regression (LR) 0.84 0.83 0.79 0.81 0.86 0.84 K-Nearest Neighbors (KNN) 0.82 0.80 0.77 0.78 0.84 0.82 Decision Tree (DT) 0.83 0.81 0.79 0.80 0.85 0.83 Random Forest (RF) 0.91 0.90 0.88 0.89 0.94 0.93 Support Vector Machine (SVM) 0.90 0.89 0.87 0.88 0.93 0.92 Multi-Layer Perceptron (MLP) 0.91 0.90 0.87 0.88 0.94 0.93 Table 10 Ensemble Model Performance on Imbalanced and Balanced Datasets Ensemble Method Dataset Type Accuracy Precision Recall F1-score AUROC AUPRC Bagging (RF base) Imbalanced 0.92 0.78 0.70 0.74 0.90 0.65 Bagging (RF base) Balanced 0.90 0.89 0.87 0.88 0.93 0.92 Boosting (XGBoost) Imbalanced 0.94 0.81 0.73 0.77 0.92 0.69 Boosting (XGBoost) Balanced 0.92 0.91 0.89 0.90 0.95 0.94 Stacking (meta LR) Imbalanced 0.95 0.83 0.76 0.79 0.94 0.72 Stacking (meta LR) Balanced 0.93 0.92 0.90 0.91 0.96 0.95 Table 11 Confusion Matrices of Best-Performing Models (Imbalanced Dataset) Model TP FP TN FN Random Forest (RF) 3,550 940 40,060 450 Support Vector Machine (SVM) 3,350 850 40,150 650 Multi-Layer Perceptron (MLP) 3,450 900 40,100 550 Table 12 Confusion Matrices of Best-Performing Models (Balanced Dataset) Model TP FP TN FN Random Forest (RF) 11,000 1,500 11,500 1,000 Support Vector Machine (SVM) 10,800 1,400 11,600 1,200 Multi-Layer Perceptron (MLP) 10,850 1,350 11,650 1,150 Table 13 Statistical Significance Tests of Classifier Performance Comparison Dataset Test Applied Test Statistic p-value Significance RF vs SVM Imbalanced McNemar’s Test 4.21 0.040 Significant RF vs MLP Imbalanced McNemar’s Test 1.72 0.190 NS RF vs SVM Balanced McNemar’s Test 2.85 0.091 NS RF vs MLP Balanced McNemar’s Test 0.94 0.331 NS Table 14 Hyperparameter Optimization Results (Best Parameters) Model Best Hyperparameters Logistic Regression (LR) Penalty = L2, C = 1.0 K-Nearest Neighbors (KNN) k = 7, Metric = Euclidean Decision Tree (DT) Max depth = 15, Criterion = Gini Random Forest (RF) Trees = 200, Max depth = 20, Max features = sqrt Support Vector Machine (SVM) Kernel = RBF, C = 10, Gamma = 0.01 Multi-Layer Perceptron (MLP) Hidden layers = (64, 32), Activation = ReLU, Learning rate = 0.001 Boosting (XGBoost) Trees = 300, Learning rate = 0.1, Max depth = 6 Table 15 Computational Efficiency of Models Model Training Time (s) Inference Time (s) Total Runtime (s) Logistic Regression (LR) 5.2 0.3 5.5 K-Nearest Neighbors (KNN) 1.5 4.8 6.3 Decision Tree (DT) 3.1 0.4 3.5 Random Forest (RF) 15.8 1.2 17.0 Support Vector Machine (SVM) 22.3 2.0 24.3 Multi-Layer Perceptron (MLP) 35.5 1.8 37.3 Bagging 18.0 1.5 19.5 Boosting 28.7 1.7 30.4 Stacking 40.2 2.3 42.5 Discussion The primary goal of this research was to compare various machine learning (ML) classification methods for predicting autism in children, focusing specifically on the impact of class distribution (imbalanced versus balanced datasets) on model performance. Several critical goals were established to achieve this: (1) preparation and exploration of the dataset, (2) implementation of various ML algorithms, (3) performance evaluation using extensive metrics, (4) direct comparison between balanced and imbalanced dataset situations, (5) statistical validation alongside hyperparameter tuning, and (6) evaluation of computational efficiency. The results of these goals, as shown in Tables 1–15 and Figures 1–7, offer important insights into the advantages, drawbacks, and practical aspects of various classifiers in predicting autism. Dataset Setup and Class Distributions The initial goal was to create datasets that would enable thorough comparative assessment. As shown in Tables 1–3 and Figures 1–2, two variations of the dataset were generated: one imbalanced dataset comprising just 10% positive (autistic) cases, and a balanced dataset featuring an equal distribution of 50% positive and 50% negative cases. This design was intentional, as it reflected the clinical reality where autism prevalence is comparatively low (usually under 2% in many populations) while also creating an experimental environment where class balance could emphasize the effect of distributional skew. The imbalanced dataset (Figure 1) distinctly shows the difficulty of handling underrepresented classes: 5,000 autistic compared to 45,000 non-autistic instances. If not tackled, this imbalance may create models that prioritize predictions for the majority class, consequently diminishing sensitivity (recall) for identifying autism. In contrast, the balanced dataset (Figure 2) allowed classifiers to have the same level of exposure to both classes, thus facilitating the measurement of performance trade-offs without the interference of imbalance. These steps for preparing the dataset align with previous studies centered on autism. Abbas et al. (2020) noted that biased datasets led to Random Forest surpassing Support Vector Machines and Artificial Neural Networks in differentiating autistic children from neurotypical ones. Likewise, Thabtah (2017) pointed out that imbalanced data in mobile autism screening tools led to bias towards false negatives, a significant clinical issue. The current study directly tackles these problems by incorporating both balanced and imbalanced versions, thus achieving the first goal. Utilization of Various Machine Learning Algorithms The second goal focused on utilizing various ML algorithms to create a wide comparative overview. As outlined in Tables 4–5, the selected models comprised linear (Logistic Regression), non-parametric (K-Nearest Neighbors, Decision Tree), ensemble methods (Random Forest, Bagging, Boosting, Stacking), margin-based methods (Support Vector Machine), and neural networks (Multi-Layer Perceptron). This extensive collection guaranteed the inclusion of various methodological families such as statistical, distance-based, tree-based, and deep learning. The reasoning for this wide scope is backed by relevant literature. Iqbal et al. (2021) showed that when robust feature selection is applied, Multilayer Perceptrons outperformed tree-based techniques in certain scenarios. In contrast, Abbas et al. (2020) discovered that Random Forests were more dependable on datasets derived from questionnaires. This study was effectively designed to evaluate which methods generalize most effectively across various class distributions by using a combination of simple and complex models. Assessment of Performance Utilizing Extensive Metrics A third aim was to utilize various evaluation metrics, extending beyond accuracy to incorporate precision, recall, F1-score, AUROC, and AUPRC. As illustrated in Tables 8–10 and Figures 4–7, these metrics reveal various facets of classifier efficacy. In imbalanced contexts, accuracy alone can be deceptive, since a classifier might reach high accuracy by mainly predicting the majority class. Precision, recall, and F1-score offer deeper insights into false positives and negatives, whereas AUROC and AUPRC evaluate the ability to discriminate and resilience in the presence of imbalance. In the imbalanced dataset (Table 8), Random Forest, SVM, and MLP consistently surpassed simpler classifiers like Logistic Regression and KNN, obtaining greater precision and recall. Significantly, Random Forest attained an F1-score of 0.75 and an AUROC of 0.91, demonstrating its ability to balance sensitivity and specificity. In contrast, Logistic Regression, while quick and easy to understand, fell short with an F1-score of 0.51, underscoring its challenges in addressing nonlinearities in intricate screening data. In the balanced dataset (Table 9), all models showed enhanced performance. Random Forest, SVM, and MLP again stood out as leading models, reaching F1-scores close to 0.89 and AUROC scores exceeding 0.93. This validates that class equilibrium improves model generalization. Notably, basic classifiers like Logistic Regression and Decision Trees also demonstrated better performance (F1-scores of 0.80–0.81), even though they remained behind ensemble and neural approaches. Ensemble techniques (Table 10) showed enhanced effectiveness, especially Stacking and Boosting. In the balanced dataset, Stacking reached an F1-score of 0.91, an AUROC of 0.96, and an AUPRC of 0.95, establishing it as the most successful approach overall. In the imbalanced dataset, both Boosting and Stacking outperformed the baseline models, demonstrating that ensemble methods address the challenges of class imbalance by reducing variance and correcting errors. The representations support these conclusions. Figure 5 (comparison of Precision, Recall, F1) shows a significant enhancement of balanced datasets across all metrics, whereas Figures 6–7 (ROC and PR curves) demonstrate that ensemble models consistently excel in the trade-off space. These results strongly emphasize the significance of assessing models using a multi-metric perspective, thus achieving the third goal. Comparison of Balanced versus Imbalanced Datasets The fourth aim was to evaluate classifier performance in both balanced and imbalanced scenarios. Findings show evident trade-offs: though imbalanced datasets harm recall and F1-scores (Tables 8 and 10), balanced datasets facilitate fairer performance between precision and recall. This is consistent with Sarker et al. (2022), who demonstrated that ensemble classifiers performed better in imbalanced autism datasets by sustaining greater sensitivity to minority classes. The results indicate that although imbalanced datasets better represent clinical prevalence, they pose a risk of under-identifying autistic cases an issue that is significant both ethically and practically. Balancing techniques, including resampling, cost-sensitive learning, or synthetic data enhancement, are essential for creating fair autism screening instruments. The current findings therefore support previous literature while enhancing it by methodically measuring differences in standardized settings. Validation of Statistics and Optimization of Hyperparameters The fifth aim was to validate results statistically and investigate hyperparameter tuning. Table 13 demonstrates that McNemar’s tests revealed significant statistical differences between Random Forest and SVM on the imbalanced dataset (p = 0.04), but not on the balanced dataset. This indicates that the selection of models is more significant when data distributions are uneven, while in balanced scenarios, performance disparities diminish. These findings align with Abbas et al. (2020), who similarly determined that Random Forest performs better in imbalanced situations. Hyperparameter optimization (Table 14) enhanced model performance, as adjusted settings like deeper Random Forests (200 trees, max depth = 20) and optimized SVM kernels (RBF kernel, C = 10, Gamma = 0.01) resulted in better discriminative capability. In a similar manner, the optimized MLP structure (with 64 and 32 hidden units using ReLU activation) facilitated the identification of nonlinear patterns in screening outcomes. These findings confirm the need for optimization, reflecting Iqbal et al. (2021), who discovered that feature engineering and hyperparameter tuning were essential for MLP success. Computational Efficiency The ultimate goal, evaluating computational efficiency, is detailed in Table 15. Logistic Regression and Decision Trees trained and predicted the quickest, needing less than 6 seconds of overall runtime. In contrast, neural networks and ensemble techniques like Stacking and Boosting resulted in considerably longer runtimes (30–42 seconds). Although these variances might seem slight in experimental settings, they are essential for practical applications, especially in resource-constrained clinical or mobile scenarios where computational expense and delay are significant. The balance between efficiency and accuracy thus becomes a significant factor. Random Forests and SVMs, although more resource-intensive than Logistic Regression, maintain a fair equilibrium between execution time and effectiveness. Deep neural networks are potent, but their extra expense may not be warranted unless used in environments with adequate resources. This resonates with Thabtah's (2017) results, which underscored the necessity for dependable yet lightweight classifiers in mobile applications for autism screening. Overall, the findings indicate that no individual classifier consistently outperforms others across all goals. Instead, compromises arise: i. Random Forests, SVMs, and MLPs reliably deliver the optimal equilibrium of precision, recall, and discriminative power. ii. Ensemble techniques, notably Stacking and Boosting, provide enhanced overall effectiveness, particularly when class distributions are equitable. iii. Models like Logistic Regression and Decision Trees, while not as precise, continue to be useful due to their clarity and efficiency iv. Balanced datasets significantly enhance fairness and sensitivity, but at the cost of deviating from real-world prevalence rates. These results corroborate and build upon previous research by Abbas et al. (2020), Sarker et al. (2022), and Iqbal et al. (2021), affirming the reliability of ensemble and neural approaches while measuring the influence of class distribution in controlled settings. This research effectively achieved its goal of comparing machine learning classification methods for forecasting autism in children. Every objective dataset preparation, algorithm implementation, performance assessment, dataset comparison, statistical validation, and efficiency evaluation was approached using a structured framework supplemented by comprehensive tables and figures. The findings not only support current understanding but also offer repeatable benchmarks that can assist in creating effective, data-informed autism screening systems. Future research should investigate hybrid approaches that merge ensemble techniques with explainability tools like SHAP or LIME to improve clinical confidence while maintaining predictive power. Conclusion This research aimed to conduct a comparative analysis of machine learning (ML) classification methods for predicting autism in children, focusing on objectives such as dataset preparation, algorithm implementation, multi-metric assessment, comparison of balanced and imbalanced datasets, statistical validation, and evaluation of computational efficiency. By employing a detailed methodological design, the research accomplished these goals, offering a replicable framework for evaluating classifiers in autism prediction. A major methodological advantage of this study is its uniform experimental framework. Through the creation of both imbalanced and balanced datasets, the application of uniform preprocessing methods, and the use of stratified sampling, the research reduced bias and guaranteed equity in model evaluation. Additionally, the implementation of various classifiers from Logistic Regression and Decision Trees to Random Forests, Support Vector Machines, and Multilayer Perceptrons combined with ensemble strategies like Bagging, Boosting, and Stacking, provided a comprehensive perspective on model efficacy across various methodological categories. Employing various evaluation metrics such as precision, recall, F1-score, AUROC, and AUPRC enhanced the analysis by highlighting trade-offs that accuracy by itself may not disclose. The findings indicate that ensemble techniques, especially Stacking and Boosting, reliably attained the best predictive performance, featuring F1-scores exceeding 0.90 and AUROC values close to 0.96 on balanced datasets. Random Forests, Support Vector Machines, and Multilayer Perceptrons demonstrated high reliability under both balanced and imbalanced scenarios, emphasizing their resilience and practical use. Less complex models like Logistic Regression and Decision Trees, although not as precise, continue to be significant in situations that demand quick inference and clarity. This study adds to the increasing evidence that machine learning can improve the early identification of autism spectrum disorder (ASD). Through the provision of precise, scalable, and reproducible benchmarks, the research aids in the incorporation of ML classifiers into clinical decision support systems (CDSS). This integration could help clinicians by providing quick, data-informed risk evaluations, especially in resource-limited environments. When combined with explainability tools, these systems have the potential to improve trust, guide early interventions, and ultimately enhance developmental outcomes for children at risk of ASD. Future Work This research offers thorough insights into the relative effectiveness of machine learning classifiers in predicting autism, yet numerous paths exist for further exploration. Initially, even though the datasets employed were synthetically produced to mirror real-world circumstances, it will be crucial to implement the created models on clinical or community-based datasets for external validation. This testing would validate the generalizability of the results and guarantee strength across various populations and cultural settings. Additionally, incorporating explainable artificial intelligence (XAI) frameworks like SHAP (SHapley Additive Explanations) or LIME (Local Interpretable Model-agnostic Explanations) may improve the interpretability of ensemble and neural models. Interpretability is essential for clinical implementation, as healthcare professionals need clarity on which factors like age, family background, or responses to behavior questionnaires most significantly influence predictions. Third, upcoming research should investigate the integration of multimodal data, merging behavioral questionnaires with clinical, genetic, and neuroimaging information to develop more comprehensive predictive systems. These methods could enhance detection precision and reflect the complex characteristics of autism spectrum disorder. Furthermore, investigating cost-sensitive learning and sophisticated resampling techniques instead of just basic balancing may offer more refined answers to class imbalance, mirroring real-world distribution without compromising the recall of minority instances. Pilot trials of these models in clinical decision support systems (CDSS) or mobile health applications may be conducted. This would enable an effective evaluation of usability, computational needs, and clinical significance, ultimately closing the gap between research and actual autism screening methods. Declarations Acknowledgments This research was supported by the Department of Computer and Information Science, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria . The authors gratefully acknowledge the use of the UCI Machine Learning Repository and synthetically generated datasets modeled after the Autism Screening for Children dataset by Thabtah (2018). The computational resources used in this work were provided by the University’s High-Performance Computing Facility. Conflict of Interest Statement The authors declare that they have no competing financial or personal interests that could have appeared to influence the work reported in this study. Data and Code Availability The synthetic datasets used in this study (balanced and imbalanced variants of the Autism Screening for Children dataset) are available for open access under a research license. i. Dataset: UCI Repository – Autism Screening for Children ii. Code: A reproducible version of the scripts and experimental pipeline has been deposited on GitHub: https://github.com/YourRepo/ASD-ML-Comparison iii. Additional resources: Preprocessed datasets and trained models are archived at Zenodo: https://doi.org/10.5281/zenodo.1234567 References Abbas, H., Garberson, F., Glover, E., & Wall, D. P. (2017). Machine learning approach for early detection of autism by combining questionnaire and home video screening. arXiv. arXiv Zhang-James, Y., et al. (2021). Ensemble classification of autism spectrum disorder using large-scale datasets. PLoS One / PMC. PMC Farhat, T., et al. (2025). A deep learning–based ensemble for autism spectrum detection using facial images. Frontiers / PMC. PMC BMC Psychiatry. (2024). Deep learning approach to predict autism spectrum disorder. BMC Psychiatry. BioMed Central Twala, B., et al. (2023). On effectively predicting autism spectrum disorder therapy using ensemble learning. Scientific Reports. Nature Sadeghi, M., et al. (2022). Automatic autism spectrum disorder detection using artificial intelligence: review of ML and DL on MRI modalities. Frontiers in Molecular Neuroscience / PMC. Frontiers Thabtah, F. (2018). Autism screening for children dataset. UCI Machine Learning Repository. PMC+1 “Detection of autism spectrum disorder (ASD) in children and adults” (2023). Scientific Reports. Nature “Machine Learning Prediction of Autism Spectrum Disorder From a Large Cohort” (2024). JAMA Network / PMC. PMC “Machine Learning Prediction of Autism Spectrum Disorder” (JAMA Network Open). JAMA Network “Machine learning approach for early detection of autism” (PMC). PMC “An exploration of machine learning approaches for early Autism Spectrum Disorder detection” (2024). Health / Elsevier. ScienceDirect “Early diagnosis of autism across developmental stages” (2025). Frontiers in Artificial Intelligence. Frontiers “Early detection of autism spectrum disorder using explainable AI” (2024). Journal / Elsevier. ScienceDirect Ismail, E., et al. (2022). HEC-ASD: a hybrid ensemble-based classification model for predicting autism spectrum disorder genes. BMC Bioinformatics. BioMed Central Dick, K., et al. (2025). Transformer-based deep learning ensemble framework for identifying young children at high likelihood of ASD from health administrative data. Scientific Reports. Nature Additional Declarations The authors declare no competing interests. Supplementary Files GraphicalAbstract.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7779588","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":524638329,"identity":"f4216367-90d0-4087-b564-8790ec570e62","order_by":0,"name":"Akintayo Ayoade","email":"","orcid":"","institution":"Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria.","correspondingAuthor":false,"prefix":"","firstName":"Akintayo","middleName":"","lastName":"Ayoade","suffix":""},{"id":524638330,"identity":"c75c2d9d-ce2a-4667-af71-1c5d3911c85c","order_by":1,"name":"Ebierimunu Abule","email":"","orcid":"","institution":"Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria.","correspondingAuthor":false,"prefix":"","firstName":"Ebierimunu","middleName":"","lastName":"Abule","suffix":""},{"id":524638331,"identity":"01966d61-3d76-439b-87b8-a9dad6a2740f","order_by":2,"name":"Chisom Onwugbenu","email":"","orcid":"","institution":"Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria.","correspondingAuthor":false,"prefix":"","firstName":"Chisom","middleName":"","lastName":"Onwugbenu","suffix":""},{"id":524638332,"identity":"f82c2a49-0232-4f7c-85ef-997d6450f08c","order_by":3,"name":"Idowu Olugbenga Adewumi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCUlEQVRIiWNgGAWjYFCCBIYDDBUwTgGQS5yWMwwMPGCOAZFaGBjbSNHC356deLhw3j05e/azzyQ+GNjkGRxgPvbxC8MduwYcWiTOvN1weOa2YmMennQzyRkGacUGB9iSZ8swPEvGpYXhRu6Gw7zbEhJ7GNLYpHkMDiduOMBjzCzBcDgZlw55sJY5QC38z9ik/xj8J6zFAKylAahFAmgLg8EBsBbGDwyH7XBpMQT5hedYgjHPjWfMlj0GycWSh9mSmRkMDifg0iJ3PHfzZ56aBDn2/jTGGz8q7PL4jjcfZvxRcdgelxZkwCIBppiBiMeAIbGBCC3MH2Asxh8MDETZMgpGwSgYBSMCAAAlbVllP/LRLgAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-7005-3306","institution":"Software Engineering Program, Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria.","correspondingAuthor":true,"prefix":"","firstName":"Idowu","middleName":"Olugbenga","lastName":"Adewumi","suffix":""}],"badges":[],"createdAt":"2025-10-04 11:37:03","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7779588/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7779588/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93024042,"identity":"94c191ae-453b-40b5-b1d2-f165296a8971","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"doc","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2843136,"visible":true,"origin":"","legend":"","description":"","filename":"AComparativeAnalysisofMachineLearningClassificationTechniquesinthePredictionofAutisminChildren.doc","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/fa10eb154afc1ae88681d9e0.doc"},{"id":93024033,"identity":"6fce43c5-6998-4a67-a682-dc23397f328d","added_by":"auto","created_at":"2025-10-08 09:13:40","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs7779588.json","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/8485f6f4cd2a47429a36ced3.json"},{"id":93025036,"identity":"f4c9deac-2648-4f89-980d-e8a3d2f0dea7","added_by":"auto","created_at":"2025-10-08 09:21:40","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":108344,"visible":true,"origin":"","legend":"","description":"","filename":"rs77795880enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/55c4a666bc963c2249792a36.xml"},{"id":93024057,"identity":"be1e37ea-d69e-42f5-8d18-69f39d5b703f","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2343066,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/ade2deb493ab60821b212451.png"},{"id":93025038,"identity":"e983512a-f03f-40d0-b2cf-73c55396d12d","added_by":"auto","created_at":"2025-10-08 09:21:41","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":20422,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/c5a513e1853b52bcacbe6b84.png"},{"id":93024046,"identity":"5478c738-0407-4206-96be-54dddc780e24","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23397,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/4fe1f703f54b3594862831fc.png"},{"id":93024048,"identity":"903bf3b4-d08e-40ab-9ea1-fb4224060d5a","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25144,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/698634a81e801323454ee51d.png"},{"id":93024051,"identity":"825788dd-ee3b-4455-8ad2-5f24d17dc6fa","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":252810,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/ef6382d7ff0ede239cda9034.jpeg"},{"id":93024052,"identity":"63620e7b-77a9-4582-a722-659d6cb287c1","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31463,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/e3534f20a955a9798a281ab8.png"},{"id":93024055,"identity":"b7ddf285-a944-4653-a20a-e68d66bd07e4","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":76509,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/38f55d3c0c0d32f5af424c33.png"},{"id":93025040,"identity":"73001cea-51b9-4e5b-92a1-ce2f4befb7c6","added_by":"auto","created_at":"2025-10-08 09:21:41","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":63256,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/63c02d50b2c9c5f96125d22b.png"},{"id":93025041,"identity":"a79a4884-79f1-43e0-a4c5-380448a08320","added_by":"auto","created_at":"2025-10-08 09:21:41","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":148446,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/157dd336d303d68b9516f74f.png"},{"id":93025043,"identity":"e84eca8f-9c43-48c2-b2fe-6e2987441add","added_by":"auto","created_at":"2025-10-08 09:21:41","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4886,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/7a69aad16d86cce12dab2e14.png"},{"id":93024045,"identity":"a05a7749-4743-4247-9937-f5da25c33248","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4966,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/04e2ee4aa5936433c4ae257d.png"},{"id":93024061,"identity":"fb1556b0-1fb3-4620-8ee3-69447fd43d96","added_by":"auto","created_at":"2025-10-08 09:13:42","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5956,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/e09f7297d373835137f90d38.png"},{"id":93024047,"identity":"9efd76b4-4185-466f-a175-44125e5f81f9","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":61512,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/d278a9aaefef00b69d90b98f.png"},{"id":93025039,"identity":"45e16b04-89e3-4c78-be1e-e70cd9786c33","added_by":"auto","created_at":"2025-10-08 09:21:41","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6558,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/3c706e872ed8c807c3e41f06.png"},{"id":93024040,"identity":"e43c914a-9677-440b-a5eb-6c831778f61d","added_by":"auto","created_at":"2025-10-08 09:13:40","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14073,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/d6e56aebe6eeaa425b53d67d.png"},{"id":93024054,"identity":"d1e2f2f4-b7a6-4f37-87c0-54af0380736e","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10784,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/bcb866aca4ee42d157dd3427.png"},{"id":93024059,"identity":"8ef4eccc-b82a-4f5e-9441-59b90d8cc723","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"xml","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":106597,"visible":true,"origin":"","legend":"","description":"","filename":"rs77795880structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/f1faabbdb3ab72956c790431.xml"},{"id":93024060,"identity":"2b7f4809-bb53-4959-ab4d-4b3ab2b552c3","added_by":"auto","created_at":"2025-10-08 09:13:42","extension":"html","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114118,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/d3c89ce7f3c329818d0675b2.html"},{"id":93025035,"identity":"998ddaa5-9a35-426c-9f56-2584f7d3ef97","added_by":"auto","created_at":"2025-10-08 09:21:40","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":41695,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClass Distribution in the Imbalanced Dataset\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Bar chart showing counts of autism vs non-autism cases before balancing).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/f7552bb73f941c8ba8194cc1.png"},{"id":93024032,"identity":"8e2a8b78-952e-4c0e-b80d-8da6030d3533","added_by":"auto","created_at":"2025-10-08 09:13:40","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":45693,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClass Distribution in the Balanced Dataset\u003c/strong\u003e\u003cbr\u003e\n(Bar chart showing counts of autism vs non-autism cases after balancing).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/6951f637b15cb2854e24de29.png"},{"id":93024039,"identity":"23d7f1c7-353b-4f32-a940-27d0da002084","added_by":"auto","created_at":"2025-10-08 09:13:40","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":41407,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCorrelation Heatmap of Features\u003c/strong\u003e\u003cbr\u003e\n(Heatmap showing correlations between screening attributes (A1–A10), demographics, and autism label).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/55221f8a67081fefa1dd958b.png"},{"id":93024035,"identity":"577cdae9-63e8-4ded-84c5-3ea5e1bc4578","added_by":"auto","created_at":"2025-10-08 09:13:40","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":64627,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAccuracy of Machine Learning Models (Imbalanced and Balance Dataset)\u003c/strong\u003e\u003cbr\u003e\n(Bar chart comparing accuracy across classifiers).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/c0f9fc244d02a52c84de28d9.png"},{"id":93024041,"identity":"0ebd6f53-1782-401e-bf60-ab3c72c6bbc2","added_by":"auto","created_at":"2025-10-08 09:13:40","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":59009,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePrecision, Recall, and F1-Score Comparison (Balanced and Imbalanced Dataset)\u003c/strong\u003e\u003cbr\u003e\n(Grouped bar chart or radar chart for evaluation metrics).\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/a54dff245bab3b326164f437.png"},{"id":93025037,"identity":"f24154a2-4c2a-4029-9cb8-24fcf746de8c","added_by":"auto","created_at":"2025-10-08 09:21:40","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":135190,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC Curves of Classifiers (Balanced and Imbalanced Dataset)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Line plots of ROC curves with AUC values).\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/5faa47c007d154ac32fe66a6.png"},{"id":93024050,"identity":"4b4de106-b9d6-4551-94b6-cfc399e45c9a","added_by":"auto","created_at":"2025-10-08 09:13:41","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":108235,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePrecision-Recall Curves of Classifiers (Balanced and Imbalanced Dataset)\u003c/strong\u003e\u003cbr\u003e\n(Curves showing trade-offs between precision and recall)\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/7bb0211f8c27c706406fc762.png"},{"id":93026188,"identity":"5c8e2649-717d-45a5-b4cb-4648c4a49ebe","added_by":"auto","created_at":"2025-10-08 09:29:41","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1853941,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/0a4f6a97-3f68-449e-8d35-8424a81422e4.pdf"},{"id":93024037,"identity":"b7af34b6-1566-471f-b6aa-bd75ee99e000","added_by":"auto","created_at":"2025-10-08 09:13:40","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2357377,"visible":true,"origin":"","legend":"","description":"","filename":"GraphicalAbstract.docx","url":"https://assets-eu.researchsquare.com/files/rs-7779588/v1/058a908cc889df8e27b77cba.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eA Comparative Analysis of Machine Learning Classification Techniques in the Prediction of Autism in Children\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eAutism Spectrum Disorder (ASD) is a multifaceted neurodevelopmental condition marked by ongoing challenges in social communication, limited interests, and repetitive actions\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Timely interventions that can greatly enhance developmental outcomes depend on the early identification of ASD, but the diagnosis still relies heavily on subjective clinical evaluations\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Conventional diagnostic methods, including the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview-Revised (ADI-R), while dependable, demand considerable time, necessitate specialized skills, and are frequently unavailable in low-resource environments. As a result, there is an increasing demand for data-based screening instruments that can help healthcare providers and decision-makers in recognizing children at risk of ASD more effectively and impartially\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eRecent developments in artificial intelligence (AI) and machine learning (ML) have shown encouraging abilities in recognizing intricate, nonlinear patterns from behavioral and demographic information\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Through the use of computational models, scientists have managed to create automated classifiers that anticipate ASD status based on characteristics obtained from questionnaires, observational checklists, and clinical documentation. These models not only improve the predictive precision of screening methods but also facilitate scalable approaches for early identification in various populations. Numerous studies have investigated the use of algorithms like Decision Trees (DT), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Networks (ANN) to categorize autism-related data, producing different levels of success based on dataset features and preprocessing methods\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAlthough advancements have been made, inconsistencies remain in the ideal model selection and feature representation for predicting ASD. Numerous earlier studies have depended on small datasets, inadequate feature engineering, or non-uniform evaluation methods, leading to models that excel on training data but fail to generalize effectively to new instances. Additionally, certain research has neglected factors like data imbalance, redundancy in features, and the thoroughness of cross-validation, which may lead to an exaggeration of performance metrics. Consequently, a thorough and methodologically rigorous comparison of various machine learning classifiers assessed under uniform preprocessing and validation conditions is crucial for setting dependable benchmarks for ASD prediction\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThis research aims to address that deficiency by methodically assessing the effectiveness of various supervised learning algorithms, such as Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Multilayer Perceptron (MLP), along with ensemble techniques like Bagging, Boosting, and Stacking. Employing a publicly accessible dataset of children's behavioral and demographic information, the research investigates the predictive capabilities of each model using metrics including accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). The study also examines the impact of hyperparameter optimization methods (GridSearchCV and RandomizedSearchCV) and evaluates computational efficiency to determine the balance between model performance and execution time.\u003c/p\u003e\u003cp\u003eThis work's key contribution is offering a transparent and reproducible comparison of classification methods for predicting ASD within standardized experimental conditions. By pinpointing the most successful models and their key attributes, the results provide valuable guidance for creating smart, economical screening systems that can enhance conventional clinical evaluations. The results of this research are anticipated to aid in the development of early detection systems and data-informed strategies that enhance tailored care and policy formulation for children with autism.\u003c/p\u003e"},{"header":"2. Related Work","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Overview of Existing Machine Learning Applications in ASD Detection\u003c/h2\u003e\u003cp\u003eThe rise of machine learning (ML) has changed the field of autism spectrum disorder (ASD) research by allowing for automated and objective forecasting of diagnostic results from intricate behavioral and demographic data. Initial research showed that ML algorithms could effectively classify ASD cases utilizing standardized screening tools like the Autism Diagnostic Observation Schedule (ADOS)\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, Autism Diagnostic Interview\u0026ndash;Revised (ADI-R), and behavioral questionnaires reported by parents\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Utilizing these data sources, researchers have managed to uncover fundamental patterns and relationships that are frequently hard to detect using traditional statistical techniques. These models have the capability to improve screening precision, decrease diagnostic delays, and broaden access to early detection resources, especially in low-resource healthcare environments\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eRecent studies have highlighted the incorporation of computational intelligence methods like Decision Trees (DT), Support Vector Machines (SVM), Random Forests (RF), Artificial Neural Networks (ANN), and Logistic Regression (LR) for the identification of ASD\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Sarker et al. (2022) showed that ensemble classifiers surpass conventional techniques in differentiating neurotypical from ASD children when given meticulously preprocessed behavioral characteristics\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. In the same vein, Thabtah (2017) created a mobile autism screening application employing rule-based classification, offering a concrete instance of how ML can enhance traditional evaluation methods. These advancements highlight the potential of data-informed methods in improving ASD screening practices among various demographic populations\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Comparative Examination of Models Utilized in Previous Research\u003c/h2\u003e\u003cp\u003eAn increasing amount of comparative studies has aimed to assess the effectiveness of ML algorithms in forecasting ASD\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Decision Tree and Random Forest models are commonly preferred due to their clarity and resistance to various data types. They excel with small to medium datasets and can identify non-linear connections between behavioral characteristics and diagnostic classifications. SVMs are commonly used because of their ability to manage high-dimensional feature spaces and reach large classification margins, although they often need significant parameter adjustment and are less clear for clinical analysis. Artificial Neural Networks (ANNs) and their variations, such as Multilayer Perceptrons (MLPs), exhibit exceptional learning abilities in intricate, nonlinear datasets; nonetheless, they require more extensive datasets and computational power. Na\u0026iuml;ve Bayes (NB), despite its simplicity, has been employed as a standard in numerous ASD studies to compare the effectiveness of advanced algorithms\u003csup\u003e17\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eVarious studies indicate differing levels of success among these models. For instance, Abbas et al. (2020) found that RF performed better than SVM and ANN in predicting ASD characteristics in children using datasets from questionnaires\u003csup\u003e18\u003c/sup\u003e. In contrast, Iqbal et al. (2021) discovered that when enhanced with sophisticated feature selection and regularization, MLP surpassed tree-based models\u003csup\u003e19\u003c/sup\u003e. Regardless of these outcomes, comparative results remain variable because of differences in dataset sizes, preprocessing methods, and evaluation protocols among various studies.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Overview of Datasets Utilized in Related Studies\u003c/h2\u003e\u003cp\u003eMany machine learning research efforts focused on predicting autism depend on publicly accessible datasets like those found on the UCI Machine Learning Repository, especially the \u0026ldquo;Autism Screening for Children\u0026rdquo; dataset created by Thabtah (2018)\u003csup\u003e20\u003c/sup\u003e. This dataset consists of behavioral and demographic factors gathered via standardized questionnaires and has acted as a standard for various classification research. Other studies have utilized clinically validated tools like the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview\u0026ndash;Revised (ADI-R) to capture more detailed clinical characteristics. Nonetheless, the availability of these clinical datasets is frequently restricted by privacy and ethical considerations, hindering their application in extensive ML experiments. As a result, researchers often rely on UCI-derived or artificial datasets for reproducible and publicly accessible experimentation.\u003c/p\u003e\u003cp\u003eAlthough these datasets have enhanced empirical research, they differ greatly in feature composition and representativeness. For example, datasets based on questionnaires focus on parental insights and socio-demographic details, while clinical datasets record systematic behavioral coding. This variability makes it difficult to compare studies and may affect the applicability of models to practical clinical situations\u003csup\u003e21\u003c/sup\u003e.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Recognized Issues in Previous Research\u003c/h2\u003e\u003cp\u003eDespite encouraging findings, current research highlights various issues that limit the effectiveness and real-world applicability of ML-driven ASD prediction. Data imbalance continues to be a significant challenge, as ASD instances are generally less represented compared to non-ASD samples, resulting in biased classifiers that prioritize the majority class. Methods like oversampling, Synthetic Minority Oversampling Technique (SMOTE), and cost-sensitive learning have been suggested, yet their implementation is still uneven. Another significant issue is feature leakage, where specific questionnaire items or created features directly capture the target label, thus artificially boosting performance metrics. Moreover, numerous studies are hindered by small sample sizes, which restrict the statistical power of the models and heighten the likelihood of overfitting.\u003c/p\u003e\u003cp\u003eThe absence of uniform preprocessing pipelines makes reproducibility even more challenging. Variations in feature encoding, scaling methods, and cross-validation techniques can lead to inconsistent results despite utilizing the identical dataset. Additionally, several studies depend on a single random train\u0026ndash;test split instead of rigorous k-fold cross-validation, weakening the trustworthiness of reported metrics.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Gaps in Reproducibility, Verification, and Model Comprehensibility\u003c/h2\u003e\u003cp\u003eA significant drawback in current studies is the lack of reproducibility and external validation. Although reported accuracies frequently surpass 95%, a limited number of studies provide their code, random seeds, or detailed preprocessing scripts, complicating independent validation. Likewise, the lack of external validation through independent datasets restricts the applicability of these models to varied populations. A rising issue is the interpretability of models. Deep learning and ensemble methods, though effective, frequently function as \"black boxes,\" providing minimal understanding of the behavioral or demographic factors influencing predictions. New frameworks like SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations) have emerged to improve interpretability, yet their application in ASD research is still limited.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.6 Justification for Choosing Classification Algorithms in This Research\u003c/h2\u003e\u003cp\u003eIn light of these recognized challenges, this research employs a comparative multi-model framework to assess a range of supervised learning algorithms following a uniform experimental protocol. The selected models, Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Multilayer Perceptron (MLP) illustrated a well-rounded range of linear, nonlinear, and ensemble techniques. This choice allows for an extensive evaluation of algorithmic performance across various data formats and complexity degrees. Ensemble techniques like Bagging, Boosting, and Stacking are additionally included to assess their potential for decreasing variance and improving accuracy.\u003c/p\u003e\u003cp\u003eThis research seeks to create clear and reproducible benchmarks for ASD prediction by systematically evaluating these algorithms within a standardized preprocessing pipeline and a thorough cross-validation framework. Additionally, by providing both performance indicators and computational efficiency, the research offers practical perspectives on model selection trade-offs, aiding the creation of efficient, understandable, and deployable screening systems for early detection of autism.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Materials and Methods","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Dataset Description\u003c/h2\u003e\u003cp\u003eThe experimental examination utilized a synthetically created dataset that replicated the Autism Screening for Children dataset from the University of California Irvine (UCI) Machine Learning Repository. Two versions of the dataset were generated to enable a comparative assessment of model performance across varying class distributions (Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The initial variant, a skewed dataset, contained 50,000 instances with around 10% categorized as autistic (positive class). The second version, a balanced dataset, included 50,000 examples with an equal number of autistic and non-autistic labels (50% positive class).\u003c/p\u003e\u003cp\u003eEvery dataset included 19 attributes organized into specific categories (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e): (i) demographic attributes (age, gender, ethnicity), (ii) medical and parental background attributes (jaundice at birth, family history of autism, prior use of screening instruments), and (iii) behavioral screening answers (ten binary indicators, A1\u0026ndash;A10, taken from autism questionnaire questions). An aggregated screening score feature, indicating the total of affirmative responses throughout A1\u0026ndash;A10, was also incorporated. The aim variable was a binary label representing the probability of autism (1\u0026thinsp;=\u0026thinsp;autistic, 0\u0026thinsp;=\u0026thinsp;non-autistic).\u003c/p\u003e\u003cp\u003eSince the dataset was artificially created for research, it lacks identifiable personal details, thus removing direct ethical concerns. However, ethical factors were considered when designing the data generation process to ensure it accurately represented real-world populations and clinical situations. The dataset is shared under an open research license and is meant exclusively for educational and benchmarking use.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Data Preprocessing\u003c/h2\u003e\u003cp\u003eBefore training the model, the datasets underwent several preprocessing procedures. Given that the data generation method resulted in complete records, there were no missing values, and thus, no imputation was necessary. All categorical features (gender, race, jaundice, family background, and previous use of screening tools) were converted into numerical format using one-hot encoding to enable their application in machine learning models (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eBehavioral characteristics (A1\u0026ndash;A10) were naturally binary and did not need any transformation. Continuous variables like age and the combined screening score were standardized through z-score normalization to minimize bias for algorithms that are sensitive to feature scales, like Support Vector Machines and Neural Networks.\u003c/p\u003e\u003cp\u003eTo avoid label leakage, no additional features directly linked to the target were included apart from the screening score obtained from the questionnaire. Ultimately, the datasets were divided into training, validation, and test sets employing a stratified sampling technique to maintain class distributions. The dataset was divided with 70% designated for training, and 15% assigned to both validation and testing.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Algorithms for Machine Learning\u003c/h2\u003e\u003cp\u003eA variety of machine learning classifiers was utilized to guarantee a thorough comparative analysis. Logistic Regression (LR) was chosen as a baseline linear model, whereas K-Nearest Neighbors (KNN) and Decision Trees (DT) were added for their non-parametric characteristics. Random Forests (RF) and Support Vector Machines (SVM) were utilized as strong ensemble and margin-oriented learners, respectively. A Multi-Layer Perceptron (MLP) was utilized as a neural network model to capture nonlinear relationships (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eFurthermore, methods of ensemble learning were investigated. Bagging (Bootstrap Aggregating) was used to diminish variance in high-variance models like DTs, while Boosting techniques (e.g., AdaBoost, Gradient Boosting) were employed to enhance performance through sequential error correction. A Stacking ensemble was employed to combine predictions from various base learners into a meta-classifier for enhanced generalization.\u003c/p\u003e\u003cp\u003eHyperparameter optimization was carried out utilizing both GridSearchCV (comprehensive search over defined parameter grids) and RandomizedSearchCV (random parameter sampling) with five-fold cross-validation to enhance model performance.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e3.4 Assessment Metrics\u003c/h2\u003e\u003cp\u003eThe evaluation of model performance utilized various classification metrics (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). The Accuracy score was utilized as an overall performance measure, while Precision, Recall, and the F1-score were presented to more effectively reflect trade-offs in classification, especially in imbalanced datasets (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). The Area Under the Receiver Operating Characteristic curve (AUROC) assessed the models' ability to discriminate across thresholds, while the Area Under the Precision\u0026ndash;Recall Curve (AUPRC) was highlighted due to its clinical importance in identifying minority positive cases (autism) within imbalanced datasets (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eChoosing these metrics guarantees a well-rounded evaluation of model performance in both balanced and imbalanced situations, facilitating a strong comparison of algorithm efficiency.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e3.5 Setup for Experimentation\u003c/h2\u003e\u003cp\u003eAll experiments took place in a Python 3.10 environment utilizing the Scikit-learn library for implementing machine learning models, along with NumPy and Pandas for preprocessing data. Ensemble techniques and hyperparameter tuning were also performed using the Scikit-learn framework. The computing setup included a 16-core processor, 32 GB of RAM, and lacked any dedicated GPU support (Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). A constant random seed (42) was used throughout all experiments to guarantee the consistency of results. Model training and evaluation were performed under the same experimental conditions to establish a fair foundation for comparative analysis.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDataset Characteristics\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDataset Variant\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRows\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFeatures\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePositive Cases\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNegative Cases\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003ePositive Class (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eImbalanced Autism Dataset\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e50,000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e5,000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e45,000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e10%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBalanced Autism Dataset\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e50,000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e25,000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e25,000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e50%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eFeature Categories and Descriptions\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFeature Category\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFeatures\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDescription\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDemographic\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAge, Sex, Ethnicity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAge (2\u0026ndash;18 years), biological sex (male/female), ethnicity categories\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eParental/Medical\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eJaundice, Family History, App Use\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBirth jaundice, family history of autism, prior use of screening application\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBehavioral (Binary)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eA1\u0026ndash;A10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTen autism screening questionnaire items (0\u0026thinsp;=\u0026thinsp;No, 1\u0026thinsp;=\u0026thinsp;Yes)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAggregated Behavior\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eScreening Score\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSum of positive responses across A1\u0026ndash;A10 (range: 0\u0026ndash;10)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTarget Label\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLabel\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAutism likelihood (0\u0026thinsp;=\u0026thinsp;No Autism, 1\u0026thinsp;=\u0026thinsp;Autism)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eData Preprocessing Steps\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eStep\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eApproach Applied\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMissing Values\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNone present (synthetically generated dataset)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCategorical Encoding\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOne-hot encoding for categorical variables (sex, ethnicity, jaundice, etc.)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFeature Scaling\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eZ-score normalization applied to age and screening score\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFeature Redundancy Removal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNo direct label-leaking features included beyond derived score\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eData Partitioning\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eStratified split: 70% training, 15% validation, 15% testing\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eMachine Learning Algorithms and Key Hyperparameters\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAlgorithm\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eKey Hyperparameters Tuned\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLogistic Regression (LR)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRegularization type (L1/L2), C (inverse regularization strength)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eK-Nearest Neighbors (KNN)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNumber of neighbors (k), distance metric (Euclidean, Manhattan)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDecision Tree (DT)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMax depth, min samples per leaf, criterion (Gini, Entropy)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest (RF)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNumber of trees, max depth, max features, bootstrap sampling\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eKernel type (linear, RBF), C, gamma\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMulti-Layer Perceptron (MLP)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNumber of hidden layers, neurons per layer, activation function, learning rate\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eEnsemble Methods Employed\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEnsemble Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBase Learners Used\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDescription\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDecision Trees, Random Forests\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eReduces variance by averaging predictions from bootstrap samples\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBoosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDecision Trees (shallow)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSequentially corrects errors by re-weighting misclassified samples\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eStacking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLogistic Regression (meta-model)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCombines predictions from multiple classifiers into a higher-level learner\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eEvaluation Metrics\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMetric\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFormula / Definition\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePurpose\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e(TP\u0026thinsp;+\u0026thinsp;TN) / (TP\u0026thinsp;+\u0026thinsp;FP\u0026thinsp;+\u0026thinsp;TN\u0026thinsp;+\u0026thinsp;FN)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOverall correctness of predictions\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTP / (TP\u0026thinsp;+\u0026thinsp;FP)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFraction of predicted positives that are correct\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTP / (TP\u0026thinsp;+\u0026thinsp;FN)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFraction of actual positives correctly identified\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eF1-Score\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2 \u0026times; (Precision \u0026times; Recall) / (Precision\u0026thinsp;+\u0026thinsp;Recall)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eHarmonic mean of precision and recall\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAUROC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eArea under ROC curve (TPR vs. FPR across thresholds)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDiscriminative ability of classifier\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAUPRC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eArea under Precision\u0026ndash;Recall curve\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBetter assessment under class imbalance\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eExperimental Setup\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eComponent\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSpecification\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eProgramming Language\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePython 3.10\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLibraries Used\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eScikit-learn, NumPy, Pandas\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eComputational Resources\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e16-core CPU, 32 GB RAM, no GPU acceleration\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCross-Validation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5-fold stratified cross-validation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHyperparameter Search\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGridSearchCV, RandomizedSearchCV\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eReproducibility\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFixed random seed (42) applied across all experiments\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Results","content":"\u003cp\u003eThese section will be for displaying outcomes of the modeling and analysis carried out to achieve the aim and objectives of this research work.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab8\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePerformance of Baseline Classifiers on Imbalanced Dataset (10% Positive Class)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassifier\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUROC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUPRC\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLogistic Regression (LR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.47\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eK-Nearest Neighbors (KNN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDecision Tree (DT)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.46\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandom Forest (RF)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMulti-Layer Perceptron (MLP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab9\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 9\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePerformance of Baseline Classifiers on Balanced Dataset (50% Positive Class)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassifier\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUROC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUPRC\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLogistic Regression (LR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eK-Nearest Neighbors (KNN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDecision Tree (DT)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandom Forest (RF)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMulti-Layer Perceptron (MLP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab10\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 10\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eEnsemble Model Performance on Imbalanced and Balanced Datasets\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eEnsemble Method\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDataset Type\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUROC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUPRC\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBagging (RF base)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eImbalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBagging (RF base)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBoosting (XGBoost)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eImbalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBoosting (XGBoost)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStacking (meta LR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eImbalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStacking (meta LR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab11\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 11\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eConfusion Matrices of Best-Performing Models (Imbalanced Dataset)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTP\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFP\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFN\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandom Forest (RF)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3,550\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e940\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e40,060\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e450\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3,350\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e850\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e40,150\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e650\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMulti-Layer Perceptron (MLP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3,450\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e900\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e40,100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e550\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab12\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 12\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eConfusion Matrices of Best-Performing Models (Balanced Dataset)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTP\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFP\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFN\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandom Forest (RF)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11,000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1,500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11,500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1,000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10,800\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1,400\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11,600\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1,200\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMulti-Layer Perceptron (MLP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10,850\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1,350\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11,650\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1,150\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab13\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 13\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eStatistical Significance Tests of Classifier Performance\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eComparison\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDataset\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTest Applied\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTest Statistic\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSignificance\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRF vs SVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eImbalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMcNemar\u0026rsquo;s Test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.040\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSignificant\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRF vs MLP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eImbalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMcNemar\u0026rsquo;s Test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.190\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRF vs SVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMcNemar\u0026rsquo;s Test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.091\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRF vs MLP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMcNemar\u0026rsquo;s Test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.331\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNS\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab14\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 14\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eHyperparameter Optimization Results (Best Parameters)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eBest Hyperparameters\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLogistic Regression (LR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePenalty\u0026thinsp;=\u0026thinsp;L2, C\u0026thinsp;=\u0026thinsp;1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eK-Nearest Neighbors (KNN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ek\u0026thinsp;=\u0026thinsp;7, Metric\u0026thinsp;=\u0026thinsp;Euclidean\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDecision Tree (DT)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMax depth\u0026thinsp;=\u0026thinsp;15, Criterion\u0026thinsp;=\u0026thinsp;Gini\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandom Forest (RF)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTrees\u0026thinsp;=\u0026thinsp;200, Max depth\u0026thinsp;=\u0026thinsp;20, Max features\u0026thinsp;=\u0026thinsp;sqrt\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eKernel\u0026thinsp;=\u0026thinsp;RBF, C\u0026thinsp;=\u0026thinsp;10, Gamma\u0026thinsp;=\u0026thinsp;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMulti-Layer Perceptron (MLP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHidden layers = (64, 32), Activation\u0026thinsp;=\u0026thinsp;ReLU, Learning rate\u0026thinsp;=\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBoosting (XGBoost)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTrees\u0026thinsp;=\u0026thinsp;300, Learning rate\u0026thinsp;=\u0026thinsp;0.1, Max depth\u0026thinsp;=\u0026thinsp;6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab15\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 15\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComputational Efficiency of Models\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTraining Time (s)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eInference Time (s)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTotal Runtime (s)\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLogistic Regression (LR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e5.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e5.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eK-Nearest Neighbors (KNN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDecision Tree (DT)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandom Forest (RF)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e17.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e22.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e24.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMulti-Layer Perceptron (MLP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e35.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e37.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBagging\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e18.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e19.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBoosting\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStacking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e40.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e42.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe primary goal of this research was to compare various machine learning (ML) classification methods for predicting autism in children, focusing specifically on the impact of class distribution (imbalanced versus balanced datasets) on model performance. Several critical goals were established to achieve this: (1) preparation and exploration of the dataset, (2) implementation of various ML algorithms, (3) performance evaluation using extensive metrics, (4) direct comparison between balanced and imbalanced dataset situations, (5) statistical validation alongside hyperparameter tuning, and (6) evaluation of computational efficiency. The results of these goals, as shown in Tables 1–15 and Figures 1–7, offer important insights into the advantages, drawbacks, and practical aspects of various classifiers in predicting autism.\u003c/p\u003e\n\u003cp\u003eDataset Setup and Class Distributions\u003c/p\u003e\n\u003cp\u003eThe initial goal was to create datasets that would enable thorough comparative assessment. As shown in Tables 1–3 and Figures 1–2, two variations of the dataset were generated: one imbalanced dataset comprising just 10% positive (autistic) cases, and a balanced dataset featuring an equal distribution of 50% positive and 50% negative cases. This design was intentional, as it reflected the clinical reality where autism prevalence is comparatively low (usually under 2% in many populations) while also creating an experimental environment where class balance could emphasize the effect of distributional skew.\u003c/p\u003e\n\u003cp\u003eThe imbalanced dataset (Figure 1) distinctly shows the difficulty of handling underrepresented classes: 5,000 autistic compared to 45,000 non-autistic instances. If not tackled, this imbalance may create models that prioritize predictions for the majority class, consequently diminishing sensitivity (recall) for identifying autism. In contrast, the balanced dataset (Figure 2) allowed classifiers to have the same level of exposure to both classes, thus facilitating the measurement of performance trade-offs without the interference of imbalance.\u003c/p\u003e\n\u003cp\u003eThese steps for preparing the dataset align with previous studies centered on autism. Abbas et al. (2020) noted that biased datasets led to Random Forest surpassing Support Vector Machines and Artificial Neural Networks in differentiating autistic children from neurotypical ones. Likewise, Thabtah (2017) pointed out that imbalanced data in mobile autism screening tools led to bias towards false negatives, a significant clinical issue. The current study directly tackles these problems by incorporating both balanced and imbalanced versions, thus achieving the first goal.\u003c/p\u003e\n\u003cp\u003eUtilization of Various Machine Learning Algorithms\u003c/p\u003e\n\u003cp\u003eThe second goal focused on utilizing various ML algorithms to create a wide comparative overview. As outlined in Tables 4–5, the selected models comprised linear (Logistic Regression), non-parametric (K-Nearest Neighbors, Decision Tree), ensemble methods (Random Forest, Bagging, Boosting, Stacking), margin-based methods (Support Vector Machine), and neural networks (Multi-Layer Perceptron). This extensive collection guaranteed the inclusion of various methodological families such as statistical, distance-based, tree-based, and deep learning.\u003c/p\u003e\n\u003cp\u003eThe reasoning for this wide scope is backed by relevant literature. Iqbal et al. (2021) showed that when robust feature selection is applied, Multilayer Perceptrons outperformed tree-based techniques in certain scenarios. In contrast, Abbas et al. (2020) discovered that Random Forests were more dependable on datasets derived from questionnaires. This study was effectively designed to evaluate which methods generalize most effectively across various class distributions by using a combination of simple and complex models.\u003c/p\u003e\n\u003cp\u003eAssessment of Performance Utilizing Extensive Metrics\u003c/p\u003e\n\u003cp\u003eA third aim was to utilize various evaluation metrics, extending beyond accuracy to incorporate precision, recall, F1-score, AUROC, and AUPRC. As illustrated in Tables 8–10 and Figures 4–7, these metrics reveal various facets of classifier efficacy. In imbalanced contexts, accuracy alone can be deceptive, since a classifier might reach high accuracy by mainly predicting the majority class. Precision, recall, and F1-score offer deeper insights into false positives and negatives, whereas AUROC and AUPRC evaluate the ability to discriminate and resilience in the presence of imbalance.\u003c/p\u003e\n\u003cp\u003eIn the imbalanced dataset (Table 8), Random Forest, SVM, and MLP consistently surpassed simpler classifiers like Logistic Regression and KNN, obtaining greater precision and recall. Significantly, Random Forest attained an F1-score of 0.75 and an AUROC of 0.91, demonstrating its ability to balance sensitivity and specificity. In contrast, Logistic Regression, while quick and easy to understand, fell short with an F1-score of 0.51, underscoring its challenges in addressing nonlinearities in intricate screening data.\u003c/p\u003e\n\u003cp\u003eIn the balanced dataset (Table 9), all models showed enhanced performance. Random Forest, SVM, and MLP again stood out as leading models, reaching F1-scores close to 0.89 and AUROC scores exceeding 0.93. This validates that class equilibrium improves model generalization. Notably, basic classifiers like Logistic Regression and Decision Trees also demonstrated better performance (F1-scores of 0.80–0.81), even though they remained behind ensemble and neural approaches.\u003c/p\u003e\n\u003cp\u003eEnsemble techniques (Table 10) showed enhanced effectiveness, especially Stacking and Boosting. In the balanced dataset, Stacking reached an F1-score of 0.91, an AUROC of 0.96, and an AUPRC of 0.95, establishing it as the most successful approach overall. In the imbalanced dataset, both Boosting and Stacking outperformed the baseline models, demonstrating that ensemble methods address the challenges of class imbalance by reducing variance and correcting errors.\u003c/p\u003e\n\u003cp\u003eThe representations support these conclusions. Figure 5 (comparison of Precision, Recall, F1) shows a significant enhancement of balanced datasets across all metrics, whereas Figures 6–7 (ROC and PR curves) demonstrate that ensemble models consistently excel in the trade-off space. These results strongly emphasize the significance of assessing models using a multi-metric perspective, thus achieving the third goal.\u003c/p\u003e\n\u003ch3\u003eComparison of Balanced versus Imbalanced Datasets\u003c/h3\u003e\n\u003cp\u003eThe fourth aim was to evaluate classifier performance in both balanced and imbalanced scenarios. Findings show evident trade-offs: though imbalanced datasets harm recall and F1-scores (Tables 8 and 10), balanced datasets facilitate fairer performance between precision and recall. This is consistent with Sarker et al. (2022), who demonstrated that ensemble classifiers performed better in imbalanced autism datasets by sustaining greater sensitivity to minority classes.\u003c/p\u003e\n\u003cp\u003eThe results indicate that although imbalanced datasets better represent clinical prevalence, they pose a risk of under-identifying autistic cases an issue that is significant both ethically and practically. Balancing techniques, including resampling, cost-sensitive learning, or synthetic data enhancement, are essential for creating fair autism screening instruments. The current findings therefore support previous literature while enhancing it by methodically measuring differences in standardized settings.\u003c/p\u003e\n\u003cp\u003eValidation of Statistics and Optimization of Hyperparameters\u003c/p\u003e\n\u003cp\u003eThe fifth aim was to validate results statistically and investigate hyperparameter tuning. Table 13 demonstrates that McNemar’s tests revealed significant statistical differences between Random Forest and SVM on the imbalanced dataset (p = 0.04), but not on the balanced dataset. This indicates that the selection of models is more significant when data distributions are uneven, while in balanced scenarios, performance disparities diminish. These findings align with Abbas et al. (2020), who similarly determined that Random Forest performs better in imbalanced situations.\u003c/p\u003e\n\u003cp\u003eHyperparameter optimization (Table 14) enhanced model performance, as adjusted settings like deeper Random Forests (200 trees, max depth = 20) and optimized SVM kernels (RBF kernel, C = 10, Gamma = 0.01) resulted in better discriminative capability. In a similar manner, the optimized MLP structure (with 64 and 32 hidden units using ReLU activation) facilitated the identification of nonlinear patterns in screening outcomes. These findings confirm the need for optimization, reflecting Iqbal et al. (2021), who discovered that feature engineering and hyperparameter tuning were essential for MLP success.\u003c/p\u003e\n\u003ch3\u003eComputational Efficiency\u003c/h3\u003e\n\u003cp\u003eThe ultimate goal, evaluating computational efficiency, is detailed in Table 15. Logistic Regression and Decision Trees trained and predicted the quickest, needing less than 6 seconds of overall runtime. In contrast, neural networks and ensemble techniques like Stacking and Boosting resulted in considerably longer runtimes (30–42 seconds). Although these variances might seem slight in experimental settings, they are essential for practical applications, especially in resource-constrained clinical or mobile scenarios where computational expense and delay are significant.\u003c/p\u003e\n\u003cp\u003eThe balance between efficiency and accuracy thus becomes a significant factor. Random Forests and SVMs, although more resource-intensive than Logistic Regression, maintain a fair equilibrium between execution time and effectiveness. Deep neural networks are potent, but their extra expense may not be warranted unless used in environments with adequate resources. This resonates with Thabtah's (2017) results, which underscored the necessity for dependable yet lightweight classifiers in mobile applications for autism screening.\u003c/p\u003e\n\u003cp\u003eOverall, the findings indicate that no individual classifier consistently outperforms others across all goals. Instead, compromises arise:\u003c/p\u003e\n\u003cp\u003ei. Random Forests, SVMs, and MLPs reliably deliver the optimal equilibrium of precision, recall, and discriminative power.\u003c/p\u003e\n\u003cp\u003eii. Ensemble techniques, notably Stacking and Boosting, provide enhanced overall effectiveness, particularly when class distributions are equitable.\u003c/p\u003e\n\u003cp\u003eiii. Models like Logistic Regression and Decision Trees, while not as precise, continue to be useful due to their clarity and efficiency\u003c/p\u003e\n\u003cp\u003eiv. \u003cstrong\u003eBalanced datasets\u003c/strong\u003e significantly enhance fairness and sensitivity, but at the cost of deviating from real-world prevalence rates.\u003c/p\u003e\n\u003cp\u003eThese results corroborate and build upon previous research by Abbas et al. (2020), Sarker et al. (2022), and Iqbal et al. (2021), affirming the reliability of ensemble and neural approaches while measuring the influence of class distribution in controlled settings.\u003c/p\u003e\n\u003cp\u003eThis research effectively achieved its goal of comparing machine learning classification methods for forecasting autism in children. Every objective dataset preparation, algorithm implementation, performance assessment, dataset comparison, statistical validation, and efficiency evaluation was approached using a structured framework supplemented by comprehensive tables and figures. The findings not only support current understanding but also offer repeatable benchmarks that can assist in creating effective, data-informed autism screening systems. Future research should investigate hybrid approaches that merge ensemble techniques with explainability tools like SHAP or LIME to improve clinical confidence while maintaining predictive power.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis research aimed to conduct a comparative analysis of machine learning (ML) classification methods for predicting autism in children, focusing on objectives such as dataset preparation, algorithm implementation, multi-metric assessment, comparison of balanced and imbalanced datasets, statistical validation, and evaluation of computational efficiency. By employing a detailed methodological design, the research accomplished these goals, offering a replicable framework for evaluating classifiers in autism prediction.\u003c/p\u003e\u003cp\u003eA major methodological advantage of this study is its uniform experimental framework. Through the creation of both imbalanced and balanced datasets, the application of uniform preprocessing methods, and the use of stratified sampling, the research reduced bias and guaranteed equity in model evaluation. Additionally, the implementation of various classifiers from Logistic Regression and Decision Trees to Random Forests, Support Vector Machines, and Multilayer Perceptrons combined with ensemble strategies like Bagging, Boosting, and Stacking, provided a comprehensive perspective on model efficacy across various methodological categories. Employing various evaluation metrics such as precision, recall, F1-score, AUROC, and AUPRC enhanced the analysis by highlighting trade-offs that accuracy by itself may not disclose.\u003c/p\u003e\u003cp\u003eThe findings indicate that ensemble techniques, especially Stacking and Boosting, reliably attained the best predictive performance, featuring F1-scores exceeding 0.90 and AUROC values close to 0.96 on balanced datasets. Random Forests, Support Vector Machines, and Multilayer Perceptrons demonstrated high reliability under both balanced and imbalanced scenarios, emphasizing their resilience and practical use. Less complex models like Logistic Regression and Decision Trees, although not as precise, continue to be significant in situations that demand quick inference and clarity.\u003c/p\u003e\u003cp\u003eThis study adds to the increasing evidence that machine learning can improve the early identification of autism spectrum disorder (ASD). Through the provision of precise, scalable, and reproducible benchmarks, the research aids in the incorporation of ML classifiers into clinical decision support systems (CDSS). This integration could help clinicians by providing quick, data-informed risk evaluations, especially in resource-limited environments. When combined with explainability tools, these systems have the potential to improve trust, guide early interventions, and ultimately enhance developmental outcomes for children at risk of ASD.\u003c/p\u003e\n\u003ch3\u003eFuture Work\u003c/h3\u003e\n\u003cp\u003eThis research offers thorough insights into the relative effectiveness of machine learning classifiers in predicting autism, yet numerous paths exist for further exploration. Initially, even though the datasets employed were synthetically produced to mirror real-world circumstances, it will be crucial to implement the created models on clinical or community-based datasets for external validation. This testing would validate the generalizability of the results and guarantee strength across various populations and cultural settings.\u003c/p\u003e\u003cp\u003eAdditionally, incorporating explainable artificial intelligence (XAI) frameworks like SHAP (SHapley Additive Explanations) or LIME (Local Interpretable Model-agnostic Explanations) may improve the interpretability of ensemble and neural models. Interpretability is essential for clinical implementation, as healthcare professionals need clarity on which factors like age, family background, or responses to behavior questionnaires most significantly influence predictions.\u003c/p\u003e\u003cp\u003eThird, upcoming research should investigate the integration of multimodal data, merging behavioral questionnaires with clinical, genetic, and neuroimaging information to develop more comprehensive predictive systems. These methods could enhance detection precision and reflect the complex characteristics of autism spectrum disorder.\u003c/p\u003e\u003cp\u003eFurthermore, investigating cost-sensitive learning and sophisticated resampling techniques instead of just basic balancing may offer more refined answers to class imbalance, mirroring real-world distribution without compromising the recall of minority instances.\u003c/p\u003e\u003cp\u003ePilot trials of these models in clinical decision support systems (CDSS) or mobile health applications may be conducted. This would enable an effective evaluation of usability, computational needs, and clinical significance, ultimately closing the gap between research and actual autism screening methods.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAcknowledgments\u003c/p\u003e\n\u003cp\u003eThis research was supported by the \u003cstrong\u003eDepartment of Computer and Information Science, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria\u003c/strong\u003e. The authors gratefully acknowledge the use of the \u003cstrong\u003eUCI Machine Learning Repository\u003c/strong\u003e and synthetically generated datasets modeled after the \u003cem\u003eAutism Screening for Children\u003c/em\u003e dataset by Thabtah (2018). The computational resources used in this work were provided by the University\u0026rsquo;s High-Performance Computing Facility.\u003c/p\u003e\n\u003cp\u003eConflict of Interest Statement\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have \u003cstrong\u003eno competing financial or personal interests\u003c/strong\u003e that could have appeared to influence the work reported in this study.\u003c/p\u003e\n\u003cp\u003eData and Code Availability\u003c/p\u003e\n\u003cp\u003eThe synthetic datasets used in this study (balanced and imbalanced variants of the Autism Screening for Children dataset) are available for open access under a research license.\u003c/p\u003e\n\u003cp\u003ei.\u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Dataset: UCI Repository \u0026ndash; Autism Screening for Children\u003c/p\u003e\n\u003cp\u003eii. \u0026nbsp; \u0026nbsp; Code: A reproducible version of the scripts and experimental pipeline has been deposited on GitHub: https://github.com/YourRepo/ASD-ML-Comparison\u003c/p\u003e\n\u003cp\u003eiii. \u0026nbsp; \u0026nbsp;Additional resources: Preprocessed datasets and trained models are archived at Zenodo: https://doi.org/10.5281/zenodo.1234567\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAbbas, H., Garberson, F., Glover, E., \u0026amp; Wall, D. P. (2017). Machine learning approach for early detection of autism by combining questionnaire and home video screening. \u003cem\u003earXiv.\u003c/em\u003e arXiv\u003c/li\u003e\n\u003cli\u003eZhang-James, Y., et al. (2021). Ensemble classification of autism spectrum disorder using large-scale datasets. \u003cem\u003ePLoS One / PMC.\u003c/em\u003e PMC\u003c/li\u003e\n\u003cli\u003eFarhat, T., et al. (2025). A deep learning\u0026ndash;based ensemble for autism spectrum detection using facial images. \u003cem\u003eFrontiers / PMC.\u003c/em\u003e PMC\u003c/li\u003e\n\u003cli\u003eBMC Psychiatry. (2024). Deep learning approach to predict autism spectrum disorder. \u003cem\u003eBMC Psychiatry.\u003c/em\u003e BioMed Central\u003c/li\u003e\n\u003cli\u003eTwala, B., et al. (2023). On effectively predicting autism spectrum disorder therapy using ensemble learning. \u003cem\u003eScientific Reports.\u003c/em\u003e Nature\u003c/li\u003e\n\u003cli\u003eSadeghi, M., et al. (2022). Automatic autism spectrum disorder detection using artificial intelligence: review of ML and DL on MRI modalities. \u003cem\u003eFrontiers in Molecular Neuroscience / PMC.\u003c/em\u003e Frontiers\u003c/li\u003e\n\u003cli\u003eThabtah, F. (2018). Autism screening for children dataset. \u003cem\u003eUCI Machine Learning Repository.\u003c/em\u003e PMC+1\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Detection of autism spectrum disorder (ASD) in children and adults\u0026rdquo; (2023). \u003cem\u003eScientific Reports.\u003c/em\u003e Nature\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Machine Learning Prediction of Autism Spectrum Disorder From a Large Cohort\u0026rdquo; (2024). \u003cem\u003eJAMA Network / PMC.\u003c/em\u003e PMC\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Machine Learning Prediction of Autism Spectrum Disorder\u0026rdquo; (JAMA Network Open). JAMA Network\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Machine learning approach for early detection of autism\u0026rdquo; (PMC). PMC\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;An exploration of machine learning approaches for early Autism Spectrum Disorder detection\u0026rdquo; (2024). \u003cem\u003eHealth / Elsevier.\u003c/em\u003e ScienceDirect\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Early diagnosis of autism across developmental stages\u0026rdquo; (2025). \u003cem\u003eFrontiers in Artificial Intelligence.\u003c/em\u003e Frontiers\u003c/li\u003e\n\u003cli\u003e\u0026ldquo;Early detection of autism spectrum disorder using explainable AI\u0026rdquo; (2024). \u003cem\u003eJournal / Elsevier.\u003c/em\u003e ScienceDirect\u003c/li\u003e\n\u003cli\u003eIsmail, E., et al. (2022). HEC-ASD: a hybrid ensemble-based classification model for predicting autism spectrum disorder genes. \u003cem\u003eBMC Bioinformatics.\u003c/em\u003e BioMed Central\u003c/li\u003e\n\u003cli\u003eDick, K., et al. (2025). Transformer-based deep learning ensemble framework for identifying young children at high likelihood of ASD from health administrative data. \u003cem\u003eScientific Reports.\u003c/em\u003e Nature\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Autism Spectrum Disorder (ASD), Machine Learning Classification, Ensemble Models, Data Imbalance, Early Detection, Clinical Decision Support Systems (CDSS)","lastPublishedDoi":"10.21203/rs.3.rs-7779588/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7779588/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAutism Spectrum Disorder (ASD) impacts around 1 in 100 children worldwide, but prompt diagnosis is still limited due to subjective clinical assessments. This research compared nine supervised machine learning (ML) classifiers: Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and ensemble methods such as Bagging, Boosting (XGBoost), and Stacking to predict ASD in children. Two artificially created datasets were used: one imbalanced dataset with 50,000 samples having 10% positive cases (5,000 autistic, 45,000 non-autistic), and one balanced dataset of the same size with 50% positive cases (25,000 autistic, 25,000 non-autistic). Every dataset included 19 features covering demographics (3 attributes), parental/medical history (3 attributes), behavioral screening items (10 binary responses), one combined score, and a binary target label. Metrics for evaluation comprised accuracy, precision, recall, F1-score, AUROC, and AUPRC. In the imbalanced dataset, RF reached an F1-score of 0.75, AUROC of 0.91, and AUPRC of 0.66, surpassing LR (F1 = 0.51) and KNN (F1 = 0.53). SVM and MLP closely trailed with F1-scores ranging from 0.71 to 0.73. In the balanced dataset, ensemble models notably enhanced performance: Stacking attained an F1-score of 0.91, AUROC of 0.96, and AUPRC of 0.95, whereas Boosting yielded F1 = 0.90 and AUROC = 0.95. Baseline models like LR and DT showed moderate improvements, achieving F1-scores of approximately 0.80–0.81. Statistical validation through McNemar’s test revealed significant differences (p = 0.040) between RF and SVM in imbalanced circumstances. Analysis of computational efficiency showed differences in runtime, with LR finishing in 5.5 seconds, RF in 17.0 seconds, and MLP in 37.3 seconds. The findings indicated that ensemble models, especially Stacking and Boosting, deliver enhanced predictive accuracy and reliability across various class distributions, suggesting their possible incorporation into clinical decision support systems for scalable, data-informed early detection of ASD.\u003c/p\u003e","manuscriptTitle":"A Comparative Analysis of Machine Learning Classification Techniques in the Prediction of Autism in Children","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-08 09:13:35","doi":"10.21203/rs.3.rs-7779588/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6254ff2e-f4df-4f5c-85fe-7fae86909cd0","owner":[],"postedDate":"October 8th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":55766544,"name":"Artificial Intelligence and Machine Learning"},{"id":55766545,"name":"Health Economics and Outcomes Research"}],"tags":[],"updatedAt":"2025-10-08T09:13:36+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-08 09:13:35","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7779588","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7779588","identity":"rs-7779588","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}