A Machine Learning Approach to Aids Clinical Trials Group (Actg) Study

A Machine Learning Approach to Aids Clinical Trials Group (Actg) Study | 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 Machine Learning Approach to Aids Clinical Trials Group (Actg) Study Lakshminarayana Rao Malyala, Mohana Krishna G, Sneha Thiyagarajan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5247011/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 AIDS - Acquired Immunodeficiency Syndrome, is a major health concern, with an estimated 39 million people living with HIV worldwide. [ 1 ] Predicting the risk of mortality in AIDS patients is important for optimizing treatment strategies and improving the outcomes. The choice of Antiretroviral Therapy (ART), whether monotherapy or combined therapy, plays a crucial role in optimizing the treatment strategies. This study aims to apply machine learning techniques to predict patient mortality within a certain window of time using the AIDS Clinical Trials Group (ACTG) Study 175 Dataset. The results demonstrate the role of Data Science and the potential of machine learning models to forecast mortality, providing valuable insights for improving the treatment. HIV/AIDS Antiretroviral Therapy (ART) Machine Learning Random Forest Classifier Class Imbalance Handling Predictive Models Figures Figure 1 Figure 2 Figure 3 I. INTRODUCTION 1.1 Background : The AIDS is a life-threatening condition caused by the HIV (Human Immunodeficiency Virus). It is a chronic disease that causes millions of casualties globally every year. HIV infection, characterized by the Human Immunodeficiency Virus, poses a significant global health challenge. While no cure exists, antiretroviral therapy (ART) has revolutionized management, transforming HIV from a debilitating and often fatal disease into a manageable chronic condition. Hence, predicting the risk of mortality in AIDS patients has become a critical aspect of treatment strategies. Antiretroviral therapy (ART) is a treatment for people with HIV. It involves taking a combination of medications targeting different stages of the viral life cycle. By suppressing viral replication and reducing viral load, ART prevents immune system deterioration and associated clinical complications. Studies have demonstrated the remarkable effectiveness of ART in prolonging life expectancy and improving quality of life for individuals living with HIV. 1.2 Significance of the Study : While advances in Antiretroviral Therapy (ART) have transformed HIV from a fatal disease to a manageable chronic condition, optimizing individual treatment strategies remains a significant challenge. In this context, machine learning (ML) emerges as a promising tool with the potential to personalize care and improve outcomes for HIV patients. A study by Daniel Niguse Mamo et al. in BMC Medical Informatics and Decision-Making utilized machine learning to predict virological failure among HIV patients on ART in Ethiopia. It demonstrated the potential of ML to optimize resource allocation and personalize care [ 2 ]. Thus, our study, involving Machine Learning strategies to predict the mortality rate helps a lot in understanding the different characteristics and the progression of HIV. The ML models delve deep into diverse patient data, unearthing the associations between specific patient characteristics and disease progression. This understanding of risk factors empowers clinicians to tailor preventative and treatment strategies for individual patients, potentially mitigating fatal outcomes. For instance, Ogbechie M et al. published research in JMIR AI applying ML to identify HIV patients at high risk of treatment interruption in Mozambique and Nigeria. Their findings highlighted the potential of ML for targeted interventions and improved patient retention [ 3 ]. Similarly, the predictive power of our models helps in pinpointing the high-risk populations and key drivers of mortality and helping in design of future clinical trials and public health interventions, accelerating progress towards more effective AIDS/HIV management. We believe our findings offer invaluable insights for resource allocation and treatment prioritization as identifying the patient groups most vulnerable to mortality ensures efficient deployment of limited healthcare resources, maximizing lives saved. An earlier study by Prosperi et al. in 2009 investigated expert rule bases, logistic regression, and non-linear ML techniques for predicting ART response. They concluded that ML models offered advantages over traditional methods in identifying complex patterns and relationships within patient data [ 4 ]. Prosperi et al. (2012) in the Journal of AIDS Reviews compared linear and non-linear statistical learning models for predicting response to ART. Their work showcased the potential of non-linear models like random forests for improved accuracy and personalized treatment decisions [ 5 ]. The EuResist project by Zazzi et al. in 2012 explored various machine learning approaches for predicting treatment response in HIV patients. They found promising results for identifying individuals at high risk of virological failure, paving the way for personalized ART regimens [ 6 ]. 1.3 Research Objectives : To develop ML models for predicting patient mortality rate. To understand the association of mortality and patient characteristics through Data Science. To explore and compare various ML algorithms like logistic regression, random forests, Xtreme Gradient Boosting. To verify whether the ML models can differentiate between patients receiving different therapies (Monotherapy and Combination therapy). II. METHODOLOGY 2.1 Data Description : The study utilizes the AIDS Clinical Trials Group Study 175 dataset. It contains healthcare statistics and categorical information about patients who have been diagnosed with AIDS. It compares monotherapy against the combination therapy and the dataset contains 2139 examples, 23 features and 1 target variable- the censoring indicator. The target variable : Censoring Indicator(cid) : Censoring (0): This occurs when the patient does not experience the event of interest (in this case, death) during the observation period. Failure (1): This occurs when the event of interest (death) happens within the observation period. Time to Event or Censoring (time) : Represents the time until an event or censoring. It is measured from the start of the study to the event of interest (in this case, death) or censoring. In the context of predicting the class of the target variable, ‘time’ provides information about how long it took for the event(death) to occur, indicating the direct measure of a patient’s survival time. A shorter time might indicate a more severe disease progression and vice versa Treatment Indicator (trt) : Indicates the type of treatment received. (0: Zidovudine (ZDV) only, 1: ZDV + ddI, 2: ZDV + Zal, 3: ddI only). ZDV in the 30 Days Prior to 175 Indicator (z30) : Indicates if the patient received ZDV in the 30 days prior to the 175th day. (0: No ZDV, 1: Received ZDV) ZDV Prior to 175 Indicator (zprior) : Indicates if the patient received ZDV prior to the 175th day. (0: No ZDV prior, 1: Received ZDV prior). Days of Pre-175 Anti-Retroviral Therapy (preanti) : No. of days the patient received anti-retroviral therapy before the 175th day. 2.2 Data Preprocessing : Analyzing the data helped in dropping some features that were having no impact on the target variable, this was achieved by calculating correlation scores between the feature and target variable and by using the Feature selection method. 2.3 Feature Selection With High dimensional data model may learn from irrelevant features and become inaccurate. The model usually performs and generalizes well when trained over the necessary features. In order to identify the most influential features, a bar graph of feature importance was generated with the help of a logistic regression model and the Permutation importance method [ 7 ]. The importance of features was determined based on the absolute value of the coefficients in the model. Features with absolute coefficient greater than 0.0015 were selected for further analysis. This threshold (0.0015) was chosen to focus on the most impactful features, in predicting the target. 2.4 Class Imbalance Handling An imbalanced dataset is said to have uneven distribution of classes. In this dataset, the target column “cid” is imbalanced, the ‘Censoring’ class has more instances as compared to the class ‘Failure’. While the model accuracy for the imbalanced data appears high, it is deceptive as it has only trained to identify the majority class and may perform poor in predicting the minority class. Not regularizing a model adequately can lead to overfitting the majority class making it perform poorly on unseen data. After evaluating various class imbalance techniques such as Under sampling , Oversampling , Class Weights method we observed optimal results using Synthetic Minority Over-sampling Technique (SMOTE). Synthetic Minority Over-Sampling (SMOTE) : SMOTE is an oversampling technique for generating synthetic samples for minority class, as described in the original paper as, ‘The minority class is over-sampled by taking each minority class sample and introducing synthetic examples along the line segments joining any/all of the k minority class nearest neighbors. Depending upon the amount of over-sampling required, neighbors from the k nearest neighbors are randomly Chosen’ [Chawla, N. V,2002]. The synthetic data point ( S ) is created by selecting a neighbor ( N ) from the K-nearest neighbors of an original data point ( O ), and the distance ( d ) between O and N is then multiplied by a random number ( r ) between 0 and 1. This process can be mathematically expressed as: S = O + r × (N – O) By generating samples with subtle variations, SMOTE reduces the chance of overfitting and ensures optimal model training, resulting in better accuracy. 2.5 Classifier Model After thorough experimentation with different machine learning models, we found that Random Forest Classifier and XGB Classifier performed the best and were our preferred choice. Random Forest Classifier (RFC) : This is an ensemble learning method that uses multiple decision trees to improve overall model performance. In this algorithm, a collection of decision trees is trained on random subsets of the data, each making independent predictions. The final prediction is obtained by combining the predictions from all the individual trees. Random Forest Classifier is very good at capturing non-linear relationships in data, making it particularly suitable for handling complex datasets where linear models might struggle. By combining predictions from multiple trees, RFC not only enhances accuracy but also effectively reduces overfitting, ensuring a more reliable performance when applied to new and unseen data. Gradient Boosting Classifier (XGB) : XGB Classifier is a powerful machine learning model that belongs to the family of gradient boosting algorithms. It constructs a robust ensemble of decision trees, each tree learning and correcting the errors of its predecessors to achieve superior predictive performance. Through a process of iterative learning and optimization, XGB Classifier excels in solving classification problems, effectively distinguishing between different categories within data. It's known for its accuracy, efficiency, and ability to handle complex datasets, making it a widely adopted choice across various domains. XGB Classifier is a gradient boosting algorithm distinct from bagging techniques like Random Forest. While both utilize ensembles of decision trees, they differ in how trees are constructed and combined for prediction: Bagging (Random Forest): Trees are trained independently on random subsets of data, with predictions averaged for a more robust model, often reducing variance and overfitting [ 8 ]. Boosting (XGB Classifier): Trees are built sequentially, each correcting errors of its predecessors, leading to a more accurate and adaptable model, often reducing bias and improving overall performance [ 9 ]. 2.5 Model Training To address the target class imbalance issue, where there were 1600 instances of one class and 400 instances of the other, we applied stratification using the stratify parameter from the scikit-learn library. This ensures that the distribution of classes in the training and testing sets remains proportionate, contributing to a better model [ 10 ]. The SMOTE based synthetic dataset was used to fit the classifiers separately, without any stratification. 2.6 Hyperparameter tuning Hyperparameter tuning is the process of finding the optimal configuration of a model's hyperparameters to maximize its performance. In RFC and XGB classifiers, this involves adjusting settings like tree depth, number of trees, and learning rate. We performed hyperparameter tuning using the Bayesian Optimized Fine-tuning, which offers an efficient approach, intelligently exploring hyperparameter combinations to identify those leading to the best results, often outperforming traditional grid search methods [ 11 ]. The hyperparameters included, and their chosen values are as follows: Random Forest Classifier (RFC) : Table 1 RFC Tuned Hyperparameters Parameter Value ccp_alpha 5.78804e-05 max_depth 15 max_leaf_nodes 26 min_impurity_decrease 0.0001463 min_samples_leaf 17 min_samples_split 15 min_weight_fraction_leaf 0.0098792 n_estimators 4100 Gradient Boosting Classifier (XGB) : Table 2 XGBC Tuned Hyperparameters Parameter Value alpha 0.268484 eta 0.641161 gamma 0.013125 lambda 0.000146 learning_rate 2.456e-05 max_depth 6 min_child_weight 0.421031 n_estimators 259 subsample 0.9878096 III. RESULTS AND DISCUSSION Model Evaluation Metrics: Assessing the Performance of the Random Forest Classifier (RFC) and Gradient Boosting Classifier At first glance, the classifiers show promising performance in the classification task. The Random Forest classifier and Gradient Boosting Classifier are evaluated on the test set, and the accuracy is determined using 5-fold cross-validation on the test set. The same approach is followed for the fine-tuned versions of these classifiers, which are fine-tuned using the Bayesian Optimized Fine-tuning algorithm. Random Forest Classifier (RFC) : Accuracy The Random Forest Classifier model exhibited an accuracy of 88.79%. Following the fine-tuning, the Random Forest Classifier yielded a modest improvement to 89.02%. This metric gave us an overall measure of the model's effectiveness, showing a decent capability to make accurate predictions. $$\:\mathbf{A}\mathbf{c}\mathbf{c}\mathbf{u}\mathbf{r}\mathbf{a}\mathbf{c}\mathbf{y}=\frac{\varvec{T}\varvec{r}\varvec{u}\varvec{e}\:\varvec{P}\varvec{o}\varvec{s}\varvec{i}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}+\varvec{T}\varvec{r}\varvec{u}\varvec{e}\:\varvec{N}\varvec{e}\varvec{g}\varvec{a}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}}{\varvec{T}\varvec{o}\varvec{t}\varvec{a}\varvec{l}\:\varvec{n}\varvec{u}\varvec{m}\varvec{b}\varvec{e}\varvec{r}\:\varvec{o}\varvec{f}\:\varvec{p}\varvec{r}\varvec{e}\varvec{d}\varvec{i}\varvec{c}\varvec{t}\varvec{i}\varvec{o}\varvec{n}\varvec{s}}$$ Precision Precision, representing the accuracy of positive predictions has given good results. The initial precision of Random Forest Classifier was 93.83%, fine-tuning the Random Forest Classifier yielded a small boost in precision of 95.06%. $$\:\mathbf{P}\mathbf{r}\mathbf{e}\mathbf{c}\mathbf{i}\mathbf{s}\mathbf{i}\mathbf{o}\mathbf{n}=\frac{\varvec{T}\varvec{r}\varvec{u}\varvec{e}\:\varvec{P}\varvec{o}\varvec{s}\varvec{i}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}}{\varvec{T}\varvec{r}\varvec{u}\varvec{e}\:\varvec{P}\varvec{o}\varvec{s}\varvec{i}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}+\varvec{F}\varvec{a}\varvec{l}\varvec{s}\varvec{e}\:\varvec{P}\varvec{o}\varvec{s}\varvec{i}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}}$$ Recall The recall metric measures the model's ability to capture all relevant instances of a class. The recall for Random Forest Classifier is 91.57%. After fine-tuning the recall slightly reduced to 90.86%, suggesting a marginal improvement in capturing the relevant instances. $$\:\mathbf{R}\mathbf{e}\mathbf{c}\mathbf{a}\mathbf{l}\mathbf{l}=\frac{\varvec{T}\varvec{r}\varvec{u}\varvec{e}\:\varvec{P}\varvec{o}\varvec{s}\varvec{i}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}}{\varvec{T}\varvec{r}\varvec{u}\varvec{e}\:\varvec{P}\varvec{o}\varvec{s}\varvec{i}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}+\varvec{F}\varvec{a}\varvec{l}\varvec{s}\varvec{e}\:\varvec{N}\varvec{e}\varvec{g}\varvec{a}\varvec{t}\varvec{i}\varvec{v}\varvec{e}\varvec{s}}$$ F1-Score F1-score is a harmonic mean of precision and recall. F1-score of Random Forest classifier started at 92.68% and improved to 92.91% after fine-tuning, indicating that the overall performance of model in terms of capturing true positives has gotten better. $$\:\mathbf{F}1\:\mathbf{S}\mathbf{c}\mathbf{o}\mathbf{r}\mathbf{e}=\frac{2\varvec{*}\varvec{P}\varvec{r}\varvec{e}\varvec{c}\varvec{i}\varvec{s}\varvec{i}\varvec{o}\varvec{n}\varvec{*}\varvec{R}\varvec{e}\varvec{c}\varvec{a}\varvec{l}\varvec{l}}{\varvec{P}\varvec{r}\varvec{e}\varvec{c}\varvec{e}\varvec{s}\varvec{i}\varvec{o}\varvec{n}+\varvec{R}\varvec{e}\varvec{c}\varvec{a}\varvec{l}\varvec{l}}$$ Gradient Boosting Classifier (XGB) : Accuracy The Gradient Boosting Classifier demonstrated an accuracy of 87.15%, and improved to 89.95% following the fine-tuning. Gradient Boosting Classifier exhibited better accuracy, indicating its potential to produce slightly more accurate predictions compared to Random Forest Classifier. Precision Gradient Boosting Classifier indicated the initial precision to be 92.90% which jumped to 95.68% after fine-tuning, meaning it now correctly flags positive instances more often. Recall The recall of XGB before tuning was 90.39% which eventually climbed to 91.45% after fine-tuning, indicating a small gain it its capacity to detect relevant instances. F1-Score Gradient Boosting Classifier demonstrated a F1-score of 91.63% and spiked up to 93.51% after fine-tuning. Gradient Boosting Classifier outperformed Random Forest Classifier in terms of overall performance indicating a measurable advantage. Table 3 Evaluation metrics of classifiers Model Accuracy Accuracy (5-fold cv) Precision Recall F1 Score Random Forest Classifier 0.8879 0.8882 0.9383 0.9157 0.9268 Random Forest Classifier (Fine Tuned) 0.8902 0.8887 0.9506 0.9086 0.9291 Gradient Boosting Classifier 0.8715 0.8845 0.9290 0.9039 0.9163 Gradient Boosting Classifier (Fine Tuned) 0.8995 0.8915 0.9568 0.9145 0.9351 Random Forest Classifier (SMOTE dataset) 0.9221 0.9119 0.9160 0.9275 0.9217 IV. CONCLUSION This study advocates the use of Machine learning models in prediction of HIV treatment outcomes with an emphasis on the use of Random-forest and XGB classifier for the same. Both the classifiers have outperformed the conventional Logistic Regression model based on the evaluation metrics. Some significant trends for better outcomes in Antiretroviral therapy regimen are identified as being Female, having longer treatment time, commitment to treatment, Higher CD4 count, prior ART history and Healthy weight. It is suggested to collect more clinical trials data and perform experiments with Machine Learning algorithms to arrive at solid conclusions for the same. Declarations Author Contribution All the authors, M.LN, MK.G, S.T have contributed equally in the coding and data analysis. The work was broken down individually and each were responsible for working on different parts of the code. M.LN worked on the data loading, data preprocessing techniques and construction of the pipeline and initializing some classifiers for classifying the processed data. MK.G worked on training the classifiers and choosing classifiers and benchmarking their performances. S.T worked on performing SMOTE analysis, Class Imbalance handling and benchmarking the models on the newer generated dataset and older dataset. MK.G has maintained the code on GitHub and is responsible for versioning. M.LN and S.T drafted the research paper, MK.G proof read from time to time and all the three authors have participated in a weekly critical review of the draft. M.LN has generated the figures using the code and cited all the references. Data Availability The experimental data that support this study is available in the UC Irvine Machine Learning Repository with DOI 10.24432/C5ZG8F and the identifier https://archive.ics.uci.edu/dataset/890/aids+clinical+trials+group+study+175 The study corresponding to this dataset and the original dataset is from The National Center for Biotechnology Information (NCBI) found at this identifier https://clinicaltrials.gov/study/NCT00000625#publications References "HIV Data WHO," WORLD HEALTH ORGANIZATION. [Online]. D. N. Mamo, "Machine learning to predict virological failure among HIV patients on antiretroviral therapy in the University of Gondar Comprehensive and Specialized Hospital, in Amhara Region, Ethiopia, 2022," BMC Medical Informatics and Decision Making, 2023. M. Ogbechie, "Predicting Treatment Interruption Among People Living With HIV in Nigeria: Machine Learning Approach," JMIR AI, vol. 2, 2023. M. Prosperi, "Investigation of expert rule bases, logistic regression, and non-linear machine learning techniques for predicting response to antiretroviral treatment.," Antivir Ther., 2009. M. C. Prosperi, "Computational Models for Prediction of Response to Antiretroviral Therapies," AIDS Reviews, vol. 14, no. 2, 2012. Z. Maurizio, "Predicting Response to Antiretroviral Treatment by Machine Learning: The EuResist Project," Intervirology, vol. 55, no. 2, 2012. A. André, "Permutation importance: a corrected feature importance measure," BIOINFORMATICS, vol. 26, no. 10, p. 1340–1347, 2010. L. Breiman, "Random Forests.," Machine Learning, vol. 45, pp. 5-32, 2001. C. G. Tianqi Chen, "XGBoost: A Scalable Tree Boosting System," The 22nd ACM SIGKDD International Conference, pp. 785-794, 2016. D. H. G. E. A. S. &. K. E. M. Tianyao Huo, "Stratified split sampling of electronic health records," BMC Medical Research Methodology, 2023. H. L. R. P. A. Jasper Snoek, "Practical Bayesian Optimization of Machine Learning Algorithms," Advances in Neural Information Processing Systems, 2012. L. Breiman, "Random Forests.," Machine Learning, vol. 45, pp. 5-32, 2001. C. G. Tianqi Chen, "XGBoost: A Scalable Tree Boosting System," The 22nd ACM SIGKDD International Conference, pp. 785-794, 2016. D. H. G. E. A. S. &. K. E. M. Tianyao Huo, "Stratified split sampling of electronic health records," BMC Medical Research Methodology, 2023. H. L. R. P. A. Jasper Snoek, "Practical Bayesian Optimization of Machine Learning Algorithms," Advances in Neural Information Processing Systems, 2012. A. Altmann, "Permutation importance: a corrected feature importance measure," BIOINFORMATICS, vol. 26, no. 10, p. 1340–1347, 2010. Additional Declarations No competing interests reported. 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-5247011","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":373064181,"identity":"05f357da-6bcd-4a6c-8957-7624acfcce41","order_by":0,"name":"Lakshminarayana Rao Malyala","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAElEQVRIie3PMUsDMRTA8ReepMuz8x0W+wmEuAhitZ8l3OCi4FhwMMdBupzOOvVbdG7I4CJ0VTIVBxeHdDvkBmMHncx1FMx/CC/wfoQApFJ/MlQAV0C8V6qFD3fe6yQsEAGDPtnS3H8R3JKM9rPzytLm2Q5wMDU6a8QZcZDKjj7mwz4C8+uL38nRk9R5LYpAjLKXd+5QI2D+MI+QhdQZCSTOykBqxwLhuBsjy5XOW3FDHJmyx7Ubd5NnqfdIWOI8EGic3IKsqpOBeCROTJlb5QqNrIr/ZVmYl/fJ9Xg4e3v1TetOZ9PK+HWEhHay75Hpzami+yH0P3PbtZxKpVL/sU8dtVKAtIo9/gAAAABJRU5ErkJggg==","orcid":"","institution":"CVR COLLEGE OF ENGINEERING, HYD","correspondingAuthor":true,"prefix":"","firstName":"Lakshminarayana","middleName":"Rao","lastName":"Malyala","suffix":""},{"id":373064182,"identity":"d5d20251-ac1d-4719-b3ac-4151bde43feb","order_by":1,"name":"Mohana Krishna G","email":"","orcid":"","institution":"CVR COLLEGE OF ENGINEERING, HYD","correspondingAuthor":false,"prefix":"","firstName":"Mohana","middleName":"Krishna","lastName":"G","suffix":""},{"id":373064183,"identity":"9c440ecf-6b21-4a6b-9c2e-5a7bfe1a668b","order_by":2,"name":"Sneha Thiyagarajan","email":"","orcid":"","institution":"CVR COLLEGE OF ENGINEERING, HYD","correspondingAuthor":false,"prefix":"","firstName":"Sneha","middleName":"","lastName":"Thiyagarajan","suffix":""}],"badges":[],"createdAt":"2024-10-11 14:53:25","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-5247011/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5247011/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":71603295,"identity":"2fd3392a-96ab-4bc8-864b-6c9efb4b244e","added_by":"auto","created_at":"2024-12-17 05:58:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":38916,"visible":true,"origin":"","legend":"\u003cp\u003ePermutation Feature Importance plot\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5247011/v1/a67eefc9c5213d3ef1e1f302.png"},{"id":71603294,"identity":"8eee88a3-04cb-40d7-963d-d5c2efaf0aa5","added_by":"auto","created_at":"2024-12-17 05:58:49","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":112564,"visible":true,"origin":"","legend":"\u003cp\u003eFeature CorrelationHeatmap\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5247011/v1/aa3d2f40b11b6c7d1757760a.png"},{"id":71603308,"identity":"26217aab-9438-4e7b-892e-4a3d354999e2","added_by":"auto","created_at":"2024-12-17 05:58:51","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":20399,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution plot of Target variable 'cid'\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCensoring\u003c/em\u003e: 1618 \u003cem\u003eFailure\u003c/em\u003e: 521\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5247011/v1/f0c226efc918e2bc6f8a4158.png"},{"id":71604650,"identity":"86582146-4233-46f3-b542-55b33ade5a95","added_by":"auto","created_at":"2024-12-17 06:06:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":714178,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5247011/v1/6225372f-ec50-40d9-afc2-b6b569ef7003.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eA Machine Learning Approach to Aids Clinical Trials Group (Actg) Study\u003c/p\u003e","fulltext":[{"header":"I. INTRODUCTION","content":"\u003cp\u003e\u003cstrong\u003e1.1 Background\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003eThe AIDS is a life-threatening condition caused by the HIV (Human Immunodeficiency Virus). It is a chronic disease that causes millions of casualties globally every year. HIV infection, characterized by the Human Immunodeficiency Virus, poses a significant global health challenge. While no cure exists, antiretroviral therapy (ART) has revolutionized management, transforming HIV from a debilitating and often fatal disease into a manageable chronic condition. Hence, predicting the risk of mortality in AIDS patients has become a critical aspect of treatment strategies. \u003cem\u003eAntiretroviral therapy (ART)\u003c/em\u003e is a treatment for people with HIV. It involves taking a combination of medications targeting different stages of the viral life cycle. By suppressing viral replication and reducing viral load, ART prevents immune system deterioration and associated clinical complications. Studies have demonstrated the remarkable effectiveness of ART in prolonging life expectancy and improving quality of life for individuals living with HIV.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.2 Significance of the Study\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003eWhile advances in Antiretroviral Therapy (ART) have transformed HIV from a fatal disease to a manageable chronic condition, optimizing individual treatment strategies remains a significant challenge. In this context, machine learning (ML) emerges as a promising tool with the potential to personalize care and improve outcomes for HIV patients. A study by Daniel Niguse Mamo et al. in BMC Medical Informatics and Decision-Making utilized machine learning to predict virological failure among HIV patients on ART in Ethiopia. It demonstrated the potential of ML to optimize resource allocation and personalize care [\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e]. Thus, our study, involving Machine Learning strategies to predict the mortality rate helps a lot in understanding the different characteristics and the progression of HIV. The ML models delve deep into diverse patient data, unearthing the associations between specific patient characteristics and disease progression. This understanding of risk factors empowers clinicians to tailor preventative and treatment strategies for individual patients, potentially mitigating fatal outcomes. For instance, Ogbechie M et al. published research in JMIR AI applying ML to identify HIV patients at high risk of treatment interruption in Mozambique and Nigeria. Their findings highlighted the potential of ML for targeted interventions and improved patient retention [\u003cspan class=\"CitationRef\"\u003e3\u003c/span\u003e]. Similarly, the predictive power of our models helps in pinpointing the high-risk populations and key drivers of mortality and helping in design of future clinical trials and public health interventions, accelerating progress towards more effective AIDS/HIV management.\u003c/p\u003e\n\u003cp\u003eWe believe our findings offer invaluable insights for resource allocation and treatment prioritization as identifying the patient groups most vulnerable to mortality ensures efficient deployment of limited healthcare resources, maximizing lives saved. An earlier study by Prosperi et al. in 2009 investigated expert rule bases, logistic regression, and non-linear ML techniques for predicting ART response. They concluded that ML models offered advantages over traditional methods in identifying complex patterns and relationships within patient data [\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e]. Prosperi et al. (2012) in the Journal of AIDS Reviews compared linear and non-linear statistical learning models for predicting response to ART. Their work showcased the potential of non-linear models like random forests for improved accuracy and personalized treatment decisions [\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e]. The EuResist project by Zazzi et al. in 2012 explored various machine learning approaches for predicting treatment response in HIV patients. They found promising results for identifying individuals at high risk of virological failure, paving the way for personalized ART regimens [\u003cspan class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.3 Research Objectives\u003c/strong\u003e:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eTo develop ML models for predicting patient mortality rate.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eTo understand the association of mortality and patient characteristics through Data Science.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eTo explore and compare various ML algorithms like logistic regression, random forests, Xtreme Gradient Boosting.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eTo verify whether the ML models can differentiate between patients receiving different therapies (Monotherapy and Combination therapy).\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e"},{"header":"II. METHODOLOGY","content":"\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.1 Data Description\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eThe study utilizes the AIDS Clinical Trials Group Study 175 dataset. It contains healthcare statistics and categorical information about patients who have been diagnosed with AIDS. It compares monotherapy against the combination therapy and the dataset contains 2139 examples, 23 features and 1 target variable- the censoring indicator.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe target variable\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"BoldUnderline\" class=\"BoldUnderline\" name=\"Emphasis\"\u003e\u003cstrong\u003eCensoring Indicator(cid)\u003c/strong\u003e\u003c/span\u003e\u003cstrong\u003e:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCensoring (0):\u0026nbsp;\u003c/strong\u003eThis occurs when the patient does not experience the event of interest (in this case, death) during the observation period.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFailure (1):\u0026nbsp;\u003c/strong\u003eThis occurs when the event of interest (death) happens within the observation period.\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"BoldUnderline\" class=\"BoldUnderline\" name=\"Emphasis\"\u003e\u003cstrong\u003eTime to Event or Censoring (time)\u003c/strong\u003e\u003c/span\u003e\u003cstrong\u003e:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRepresents the time until an event or censoring. It is measured from the start of the study to the event of interest (in this case, death) or censoring.\u003c/p\u003e\n\u003cp\u003eIn the context of predicting the class of the target variable, \u0026lsquo;time\u0026rsquo; provides information about how long it took for the event(death) to occur, indicating the direct measure of a patient\u0026rsquo;s survival time. A shorter time might indicate a more severe disease progression and vice versa\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"BoldUnderline\" class=\"BoldUnderline\" name=\"Emphasis\"\u003e\u003cstrong\u003eTreatment Indicator (trt)\u003c/strong\u003e\u003c/span\u003e\u003cstrong\u003e:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIndicates the type of treatment received. (0: Zidovudine (ZDV) only, 1: ZDV\u0026thinsp;+\u0026thinsp;ddI, 2: ZDV\u0026thinsp;+\u0026thinsp;Zal, 3: ddI only).\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"BoldUnderline\" class=\"BoldUnderline\" name=\"Emphasis\"\u003e\u003cstrong\u003eZDV in the 30 Days Prior to 175 Indicator (z30)\u003c/strong\u003e\u003c/span\u003e\u003cstrong\u003e:\u003c/strong\u003e Indicates if the patient received ZDV in the 30 days prior to the 175th day. (0: No ZDV, 1: Received ZDV)\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"BoldUnderline\" class=\"BoldUnderline\" name=\"Emphasis\"\u003e\u003cstrong\u003eZDV Prior to 175 Indicator (zprior)\u003c/strong\u003e\u003c/span\u003e\u003cstrong\u003e:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIndicates if the patient received ZDV prior to the 175th day. (0: No ZDV prior, 1: Received ZDV prior).\u003c/p\u003e\n\u003cp\u003e\u003cspan type=\"BoldUnderline\" class=\"BoldUnderline\" name=\"Emphasis\"\u003e\u003cstrong\u003eDays of Pre-175 Anti-Retroviral Therapy (preanti)\u003c/strong\u003e\u003c/span\u003e\u003cstrong\u003e:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo. of days the patient received anti-retroviral therapy before the 175th day.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 Data Preprocessing\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003eAnalyzing the data helped in dropping some features that were having no impact on the target variable, this was achieved by calculating correlation scores between the feature and target variable and by using the Feature selection method.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Feature Selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWith High dimensional data model may learn from irrelevant features and become inaccurate. The model usually performs and generalizes well when trained over the necessary features. In order to identify the most influential features, a bar graph of feature importance was generated with the help of a logistic regression model and the Permutation importance method [\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e]. The importance of features was determined based on the absolute value of the coefficients in the model. Features with absolute coefficient greater than 0.0015 were selected for further analysis. This threshold (0.0015) was chosen to focus on the most impactful features, in predicting the target.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4 Class Imbalance Handling\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAn imbalanced dataset is said to have uneven distribution of classes. In this dataset, the target column \u003cstrong\u003e\u0026ldquo;cid\u0026rdquo;\u003c/strong\u003e is imbalanced, the \u0026lsquo;Censoring\u0026rsquo; class has more instances as compared to the class \u0026lsquo;Failure\u0026rsquo;.\u003c/p\u003e\n\u003cp\u003eWhile the model accuracy for the imbalanced data appears high, it is deceptive as it has only trained to identify the majority class and may perform poor in predicting the minority class. Not regularizing a model adequately can lead to overfitting the majority class making it perform poorly on unseen data. After evaluating various class imbalance techniques such as \u003cstrong\u003eUnder sampling\u003c/strong\u003e, \u003cstrong\u003eOversampling\u003c/strong\u003e, \u003cstrong\u003eClass Weights\u003c/strong\u003e method we observed optimal results using Synthetic Minority Over-sampling Technique (SMOTE).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSynthetic Minority Over-Sampling (SMOTE)\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003eSMOTE is an oversampling technique for generating synthetic samples for minority class, as described in the original paper as, \u0026lsquo;The minority class is over-sampled by taking each minority class sample and introducing synthetic examples along the line segments joining any/all of the k minority class nearest neighbors. Depending upon the amount of over-sampling required, neighbors from the k nearest neighbors are randomly\u003c/p\u003e\n\u003cp\u003eChosen\u0026rsquo; [Chawla, N. V,2002].\u003c/p\u003e\n\u003cp\u003eThe synthetic data point (\u003cem\u003eS\u003c/em\u003e) is created by selecting a neighbor (\u003cem\u003eN\u003c/em\u003e) from the K-nearest neighbors of an original data point (\u003cem\u003eO\u003c/em\u003e), and the distance (\u003cem\u003ed\u003c/em\u003e) between \u003cem\u003eO\u003c/em\u003e and \u003cem\u003eN\u003c/em\u003e is then multiplied by a random number (\u003cem\u003er\u003c/em\u003e) between 0 and 1. This process can be mathematically expressed as:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eS\u0026thinsp;=\u0026thinsp;O\u0026thinsp;+\u0026thinsp;r \u0026times; (N \u0026ndash; O)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBy generating samples with subtle variations, SMOTE reduces the chance of overfitting and ensures optimal model training, resulting in better accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 Classifier Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAfter thorough experimentation with different machine learning models, we found that Random Forest Classifier and XGB Classifier performed the best and were our preferred choice.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRandom Forest Classifier (RFC)\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003eThis is an ensemble learning method that uses multiple decision trees to improve overall model performance. In this algorithm, a collection of decision trees is trained on random subsets of the data, each making independent predictions. The final prediction is obtained by combining the predictions from all the individual trees.\u003c/p\u003e\n\u003cp\u003eRandom Forest Classifier is very good at capturing non-linear relationships in data, making it particularly suitable for handling complex datasets where linear models might struggle. By combining predictions from multiple trees, RFC not only enhances accuracy but also effectively reduces overfitting, ensuring a more reliable performance when applied to new and unseen data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGradient Boosting Classifier (XGB)\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003eXGB Classifier is a powerful machine learning model that belongs to the family of gradient boosting algorithms. It constructs a robust ensemble of decision trees, each tree learning and correcting the errors of its predecessors to achieve superior predictive performance. Through a process of iterative learning and optimization, XGB Classifier excels in solving classification problems, effectively distinguishing between different categories within data. It\u0026apos;s known for its accuracy, efficiency, and ability to handle complex datasets, making it a widely adopted choice across various domains.\u003c/p\u003e\n\u003cp\u003eXGB Classifier is a gradient boosting algorithm distinct from bagging techniques like Random Forest. While both utilize ensembles of decision trees, they differ in how trees are constructed and combined for prediction:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eBagging (Random Forest): Trees are trained independently on random subsets of data, with predictions averaged for a more robust model, often reducing variance and overfitting [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eBoosting (XGB Classifier): Trees are built sequentially, each correcting errors of its predecessors, leading to a more accurate and adaptable model, often reducing bias and improving overall performance [\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 Model Training\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo address the target class imbalance issue, where there were 1600 instances of one class and 400 instances of the other, we applied stratification using the \u003cem\u003estratify\u003c/em\u003e parameter from the \u003cem\u003escikit-learn\u003c/em\u003e library. This ensures that the distribution of classes in the training and testing sets remains proportionate, contributing to a better model [\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e]. The SMOTE based synthetic dataset was used to fit the classifiers separately, without any stratification.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.6 Hyperparameter tuning\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHyperparameter tuning is the process of finding the optimal configuration of a model\u0026apos;s hyperparameters to maximize its performance. In RFC and XGB classifiers, this involves adjusting settings like tree depth, number of trees, and learning rate. We performed hyperparameter tuning using the Bayesian Optimized Fine-tuning, which offers an efficient approach, intelligently exploring hyperparameter combinations to identify those leading to the best results, often outperforming traditional grid search methods [\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eThe hyperparameters included, and their chosen values are as follows:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRandom Forest Classifier (RFC)\u003c/strong\u003e:\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eRFC Tuned Hyperparameters\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eParameter\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eValue\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\u003eccp_alpha\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.78804e-05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emax_depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emax_leaf_nodes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emin_impurity_decrease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.0001463\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emin_samples_leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emin_samples_split\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emin_weight_fraction_leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.0098792\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003en_estimators\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003eGradient Boosting Classifier (XGB)\u003c/strong\u003e:\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eXGBC Tuned Hyperparameters\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eParameter\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eValue\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\u003ealpha\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.268484\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eeta\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.641161\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003egamma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.013125\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003elambda\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.000146\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003elearning_rate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.456e-05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emax_depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emin_child_weight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.421031\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003en_estimators\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e259\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esubsample\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9878096\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"III. RESULTS AND DISCUSSION","content":"\u003cp\u003e \u003cb\u003eModel Evaluation Metrics: Assessing the Performance of the Random Forest Classifier (RFC) and Gradient Boosting Classifier\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAt first glance, the classifiers show promising performance in the classification task. The Random Forest classifier and Gradient Boosting Classifier are evaluated on the test set, and the accuracy is determined using 5-fold cross-validation on the test set. The same approach is followed for the fine-tuned versions of these classifiers, which are fine-tuned using the Bayesian Optimized Fine-tuning algorithm.\u003c/p\u003e \u003cp\u003e \u003cb\u003eRandom Forest Classifier (RFC)\u003c/b\u003e:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eAccuracy\u003c/strong\u003e \u003cp\u003eThe Random Forest Classifier model exhibited an accuracy of 88.79%. Following the fine-tuning, the Random Forest Classifier yielded a modest improvement to 89.02%. This metric gave us an overall measure of the model's effectiveness, showing a decent capability to make accurate predictions.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equa\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\mathbf{A}\\mathbf{c}\\mathbf{c}\\mathbf{u}\\mathbf{r}\\mathbf{a}\\mathbf{c}\\mathbf{y}=\\frac{\\varvec{T}\\varvec{r}\\varvec{u}\\varvec{e}\\:\\varvec{P}\\varvec{o}\\varvec{s}\\varvec{i}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}+\\varvec{T}\\varvec{r}\\varvec{u}\\varvec{e}\\:\\varvec{N}\\varvec{e}\\varvec{g}\\varvec{a}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}}{\\varvec{T}\\varvec{o}\\varvec{t}\\varvec{a}\\varvec{l}\\:\\varvec{n}\\varvec{u}\\varvec{m}\\varvec{b}\\varvec{e}\\varvec{r}\\:\\varvec{o}\\varvec{f}\\:\\varvec{p}\\varvec{r}\\varvec{e}\\varvec{d}\\varvec{i}\\varvec{c}\\varvec{t}\\varvec{i}\\varvec{o}\\varvec{n}\\varvec{s}}$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePrecision\u003c/strong\u003e \u003cp\u003ePrecision, representing the accuracy of positive predictions has given good results. The initial precision of Random Forest Classifier was 93.83%, fine-tuning the Random Forest Classifier\u003c/p\u003e \u003c/p\u003e \u003cp\u003eyielded a small boost in precision of 95.06%.\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:\\mathbf{P}\\mathbf{r}\\mathbf{e}\\mathbf{c}\\mathbf{i}\\mathbf{s}\\mathbf{i}\\mathbf{o}\\mathbf{n}=\\frac{\\varvec{T}\\varvec{r}\\varvec{u}\\varvec{e}\\:\\varvec{P}\\varvec{o}\\varvec{s}\\varvec{i}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}}{\\varvec{T}\\varvec{r}\\varvec{u}\\varvec{e}\\:\\varvec{P}\\varvec{o}\\varvec{s}\\varvec{i}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}+\\varvec{F}\\varvec{a}\\varvec{l}\\varvec{s}\\varvec{e}\\:\\varvec{P}\\varvec{o}\\varvec{s}\\varvec{i}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eRecall\u003c/strong\u003e \u003cp\u003eThe recall metric measures the model's ability to capture all relevant instances of a class. The recall for Random Forest Classifier is 91.57%. After fine-tuning the recall slightly reduced to 90.86%, suggesting a marginal improvement in capturing the relevant instances.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equc\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:\\mathbf{R}\\mathbf{e}\\mathbf{c}\\mathbf{a}\\mathbf{l}\\mathbf{l}=\\frac{\\varvec{T}\\varvec{r}\\varvec{u}\\varvec{e}\\:\\varvec{P}\\varvec{o}\\varvec{s}\\varvec{i}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}}{\\varvec{T}\\varvec{r}\\varvec{u}\\varvec{e}\\:\\varvec{P}\\varvec{o}\\varvec{s}\\varvec{i}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}+\\varvec{F}\\varvec{a}\\varvec{l}\\varvec{s}\\varvec{e}\\:\\varvec{N}\\varvec{e}\\varvec{g}\\varvec{a}\\varvec{t}\\varvec{i}\\varvec{v}\\varvec{e}\\varvec{s}}$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eF1-Score\u003c/strong\u003e \u003cp\u003eF1-score is a harmonic mean of precision and recall. F1-score of Random Forest classifier started at 92.68% and improved to 92.91% after fine-tuning, indicating that the overall performance of model in terms of capturing true positives has gotten better.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equd\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$$\\:\\mathbf{F}1\\:\\mathbf{S}\\mathbf{c}\\mathbf{o}\\mathbf{r}\\mathbf{e}=\\frac{2\\varvec{*}\\varvec{P}\\varvec{r}\\varvec{e}\\varvec{c}\\varvec{i}\\varvec{s}\\varvec{i}\\varvec{o}\\varvec{n}\\varvec{*}\\varvec{R}\\varvec{e}\\varvec{c}\\varvec{a}\\varvec{l}\\varvec{l}}{\\varvec{P}\\varvec{r}\\varvec{e}\\varvec{c}\\varvec{e}\\varvec{s}\\varvec{i}\\varvec{o}\\varvec{n}+\\varvec{R}\\varvec{e}\\varvec{c}\\varvec{a}\\varvec{l}\\varvec{l}}$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eGradient Boosting Classifier (XGB)\u003c/b\u003e:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eAccuracy\u003c/strong\u003e \u003cp\u003eThe Gradient Boosting Classifier demonstrated an accuracy of 87.15%, and improved to 89.95% following the fine-tuning.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eGradient Boosting Classifier exhibited better accuracy, indicating its potential to produce slightly more accurate predictions compared to Random Forest Classifier.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePrecision\u003c/strong\u003e \u003cp\u003eGradient Boosting Classifier indicated the initial precision to be 92.90% which jumped to 95.68% after fine-tuning, meaning it now correctly flags positive instances more often.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eRecall\u003c/strong\u003e \u003cp\u003eThe recall of XGB before tuning was 90.39% which eventually climbed to 91.45% after fine-tuning, indicating a small gain it its capacity to detect relevant instances.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eF1-Score\u003c/strong\u003e \u003cp\u003eGradient Boosting Classifier demonstrated a F1-score of 91.63% and spiked up to 93.51% after fine-tuning. Gradient Boosting Classifier outperformed Random Forest Classifier in terms of overall performance indicating a measurable advantage.\u003c/p\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\u003eEvaluation metrics of classifiers\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=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003cp\u003e(5-fold cv)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eF1 Score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest Classifier\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8879\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8882\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9383\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9157\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9268\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest Classifier (Fine Tuned)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8902\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8887\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9506\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9086\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9291\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGradient Boosting Classifier\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8715\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8845\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9290\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9039\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9163\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGradient Boosting Classifier (Fine Tuned)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8995\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9145\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9351\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest Classifier (SMOTE dataset)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9221\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9119\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9160\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9275\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9217\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"IV. CONCLUSION","content":"\u003cp\u003eThis study advocates the use of Machine learning models in prediction of HIV treatment outcomes with an emphasis on the use of Random-forest and XGB classifier for the same. Both the classifiers have outperformed the conventional Logistic Regression model based on the evaluation metrics. Some significant trends for better outcomes in Antiretroviral therapy regimen are identified as being Female, having longer treatment time, commitment to treatment, Higher CD4 count, prior ART history and Healthy weight. It is suggested to collect more clinical trials data and perform experiments with Machine Learning algorithms to arrive at solid conclusions for the same.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAll the authors, M.LN, MK.G, S.T have contributed equally in the coding and data analysis. The work was broken down individually and each were responsible for working on different parts of the code. M.LN worked on the data loading, data preprocessing techniques and construction of the pipeline and initializing some classifiers for classifying the processed data. MK.G worked on training the classifiers and choosing classifiers and benchmarking their performances. S.T worked on performing SMOTE analysis, Class Imbalance handling and benchmarking the models on the newer generated dataset and older dataset. MK.G has maintained the code on GitHub and is responsible for versioning. M.LN and S.T drafted the research paper, MK.G proof read from time to time and all the three authors have participated in a weekly critical review of the draft. M.LN has generated the figures using the code and cited all the references.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe experimental data that support this study is available in the UC Irvine Machine Learning Repository with DOI 10.24432/C5ZG8F and the identifier https://archive.ics.uci.edu/dataset/890/aids+clinical+trials+group+study+175 The study corresponding to this dataset and the original dataset is from The National Center for Biotechnology Information (NCBI) found at this identifier https://clinicaltrials.gov/study/NCT00000625#publications\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003e\"HIV Data WHO,\" WORLD HEALTH ORGANIZATION. [Online]. \u003c/li\u003e\n \u003cli\u003eD. N. Mamo, \"Machine learning to predict virological failure among HIV patients on antiretroviral therapy in the University of Gondar Comprehensive and Specialized Hospital, in Amhara Region, Ethiopia, 2022,\" \u003cem\u003eBMC Medical Informatics and Decision Making, \u003c/em\u003e2023. \u003c/li\u003e\n \u003cli\u003eM. Ogbechie, \"Predicting Treatment Interruption Among People Living With HIV in Nigeria: Machine Learning Approach,\" \u003cem\u003eJMIR AI, \u003c/em\u003evol. 2, 2023. \u003c/li\u003e\n \u003cli\u003eM. Prosperi, \"Investigation of expert rule bases, logistic regression, and non-linear machine learning techniques for predicting response to antiretroviral treatment.,\" \u003cem\u003eAntivir Ther., \u003c/em\u003e2009. \u003c/li\u003e\n \u003cli\u003eM. C. Prosperi, \"Computational Models for Prediction of Response to Antiretroviral Therapies,\" \u003cem\u003eAIDS Reviews, \u003c/em\u003evol. 14, no. 2, 2012. \u003c/li\u003e\n \u003cli\u003eZ. Maurizio, \"Predicting Response to Antiretroviral Treatment by Machine Learning: The EuResist Project,\" \u003cem\u003eIntervirology, \u003c/em\u003evol. 55, no. 2, 2012. \u003c/li\u003e\n \u003cli\u003eA. André, \"Permutation importance: a corrected feature importance measure,\" \u003cem\u003eBIOINFORMATICS, \u003c/em\u003evol. 26, no. 10, p. 1340–1347, 2010. \u003c/li\u003e\n \u003cli\u003eL. Breiman, \"Random Forests.,\" \u003cem\u003eMachine Learning, \u003c/em\u003evol. 45, pp. 5-32, 2001. \u003c/li\u003e\n \u003cli\u003eC. G. Tianqi Chen, \"XGBoost: A Scalable Tree Boosting System,\" \u003cem\u003eThe 22nd ACM SIGKDD International Conference, \u003c/em\u003epp. 785-794, 2016. \u003c/li\u003e\n \u003cli\u003eD. H. G. E. A. S. \u0026amp;. K. E. M. Tianyao Huo, \"Stratified split sampling of electronic health records,\" \u003cem\u003eBMC Medical Research Methodology, \u003c/em\u003e2023. \u003c/li\u003e\n \u003cli\u003eH. L. R. P. A. Jasper Snoek, \"Practical Bayesian Optimization of Machine Learning Algorithms,\" \u003cem\u003eAdvances in Neural Information Processing Systems, \u003c/em\u003e2012. \u003c/li\u003e\n \u003cli\u003eL. Breiman, \"Random Forests.,\" \u003cem\u003eMachine Learning, \u003c/em\u003evol. 45, pp. 5-32, 2001. \u003c/li\u003e\n \u003cli\u003eC. G. Tianqi Chen, \"XGBoost: A Scalable Tree Boosting System,\" \u003cem\u003eThe 22nd ACM SIGKDD International Conference, \u003c/em\u003epp. 785-794, 2016. \u003c/li\u003e\n \u003cli\u003eD. H. G. E. A. S. \u0026amp;. K. E. M. Tianyao Huo, \"Stratified split sampling of electronic health records,\" \u003cem\u003eBMC Medical Research Methodology, \u003c/em\u003e2023. \u003c/li\u003e\n \u003cli\u003eH. L. R. P. A. Jasper Snoek, \"Practical Bayesian Optimization of Machine Learning Algorithms,\" \u003cem\u003eAdvances in Neural Information Processing Systems, \u003c/em\u003e2012. \u003c/li\u003e\n \u003cli\u003eA. Altmann, \"Permutation importance: a corrected feature importance measure,\" \u003cem\u003eBIOINFORMATICS, \u003c/em\u003evol. 26, no. 10, p. 1340–1347, 2010. \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":true,"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":"HIV/AIDS, Antiretroviral Therapy (ART), Machine Learning, Random Forest Classifier, Class Imbalance Handling, Predictive Models","lastPublishedDoi":"10.21203/rs.3.rs-5247011/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5247011/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAIDS - Acquired Immunodeficiency Syndrome, is a major health concern, with an estimated 39\u0026nbsp;million people living with HIV worldwide. [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] Predicting the risk of mortality in AIDS patients is important for optimizing treatment strategies and improving the outcomes. The choice of Antiretroviral Therapy (ART), whether monotherapy or combined therapy, plays a crucial role in optimizing the treatment strategies. This study aims to apply machine learning techniques to predict patient mortality within a certain window of time using the AIDS Clinical Trials Group (ACTG) Study 175 Dataset. The results demonstrate the role of Data Science and the potential of machine learning models to forecast mortality, providing valuable insights for improving the treatment.\u003c/p\u003e","manuscriptTitle":"A Machine Learning Approach to Aids Clinical Trials Group (Actg) Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-17 05:58:44","doi":"10.21203/rs.3.rs-5247011/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":"19ef028b-f64f-4597-bd8c-ddfc4653ab15","owner":[],"postedDate":"December 17th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-12-17T05:58:44+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-17 05:58:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5247011","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5247011","identity":"rs-5247011","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}