Machine Learning Framework for Individualised Health Profiling Using Multimodal Physiological and Neurocognitive Biomarkers

Machine Learning Framework for Individualised Health Profiling Using Multimodal Physiological and Neurocognitive Biomarkers | 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 Article Machine Learning Framework for Individualised Health Profiling Using Multimodal Physiological and Neurocognitive Biomarkers Navdha Bhardwaj, Akash Singh, Ritik Sharma, Gitanshu Chaudhary, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7044176/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 Correct classification of individualised health profiles is paramount for the progress of precision health strategies but is difficult for conventional constitution-based systems based on subjective judgments. This work fills this void by suggesting a machine learning paradigm that objectively characterises individual health types based on multimodal physiological and neurocognitive markers. Heart rate variability (HRV), electroencephalography (EEG), gait dynamics, facial thermography, facial imaging, and cognitive-emotional responses in a virtual reality (VR) environment were obtained from 48 subjects. Feature selection, an integration of analysis of variance (ANOVA) and Random Forest-based importance ranking, minimised the dataset from 214 to 69 key variables. A novel meta ensemble model, PRAK-KNN, using multiple machine learning (ML) algorithms, XGBoost and Extra Trees for adaptive neighbor weighting, achieved an accuracy in classification of up to 76% with different feature subsets. In virtual reality-based tests, cognitive load and emotional arousal were key discriminators among individual profiles. This study illustrates an objective, replicable method for constitution-type classification, opening the door to incorporating classical health ideas into contemporary data-driven personalised medicine paradigms. Health sciences/Biomarkers Biological sciences/Computational biology and bioinformatics Physical sciences/Engineering Health sciences/Health care Physical sciences/Mathematics and computing Biological sciences/Neuroscience Personalised health profile machine learning multimodal biometrics physiological indications cognitive evaluation virtual reality environments Figures Figure 1 Figure 2 1. Introduction With the goal of matching therapeutic approaches with unique physiological, psychological, and behavioral characteristics, personalized health profiling has become a crucial component of precision medicine [ 1 ]. Constitution-based systems, such as Ayurveda's Prakriti model, have long offered systematic approaches to health classification, among the numerous ancient frameworks that suggest such individualized characterization. Based on innate characteristics such as temperament, bodily shape, and behavioral tendencies, Prakriti categorizes people into three constitution types: Vata, Pitta, and Kapha [ 2 , 3 ]. These categories affect not just health status but also treatment responses, disease susceptibility, and lifestyle management advice. However, diagnosis has historically depended on non-standardized physical observation and verbal interpretation, and Prakriti evaluations have historically depended on the subjective judgment of Ayurvedic practitioners. The incorporation of constitution-based systems into contemporary healthcare, where data-driven validation and scientific repeatability are necessary for clinical applicability, has been hampered by this dependence on subjective techniques [ 5 , 6 ]. There is renewed interest in employing objective measurements to validate traditional health systems as a result of recent technical advancements in biomedical signal processing, wearable sensors, and machine learning [ 7 – 9 ]. Using heart rate variability (HRV), researchers have tried to link Prakriti types to different autonomic profiles. They have found that Kapha types have parasympathetic dominance and Pitta types frequently demonstrate more sympathetic activity [ 10 ]. Along with facial morphological features examined using machine learning techniques, electroencephalography (EEG) and facial thermal imaging have also demonstrated potential in identifying patterns unique to a person's composition [11,12]. Additionally, Prakriti types may correlate to genetic polymorphisms associated with immunity and metabolism, according to genomic studies in Ayurgenomics [ 13 , 14 ]. Despite their significance, these findings lack the coherence necessary for a reliable and clinically scalable diagnostic tool because they are still fragmented and modality-specific. More significantly, cognitive and affective aspects that are essential to constitution typologies have been overlooked in traditional evaluations. According to Ayurvedic scriptures, Pitta types are deliberate and concentrated, Kapha types are grounded yet sluggish to change, and Vata types are creative and reactive [ 2 ]. These psychological profiles suggest that constitution types are neurocognitive as well as physical factors. However, the existing research lacks empirical attempts to measure these cognitive-affective markers, especially in ecologically realistic, real-time contexts [15,16]. Cognitive profiling provides a technique to close this crucial gap, particularly through immersive and interactive paradigms. By presenting a multimodal machine learning system called PRAK-KNN (Prakriti-aware k-Nearest Neighbour), which integrates physiological and neurocognitive data to objectively define constitution kinds, this study overcomes these constraints. Six major data streams are included in the pipeline: gait biometrics, facial morphology, facial thermal imaging, HRV, EEG, and auditory cognition in VR-based narrative activities [ 4 ]. Beyond single-modality techniques, this integrated architecture offers a scalable, reproducible approach to constitution profiling that supports the objectives of precision medicine. This study's uniqueness comes from its thorough methodological integration of several constitution profile dimensions. This study creates a unified machine learning pipeline that concurrently integrates six biometric and behavioral modalities: HRV, EEG, thermal imaging, gait analysis, facial morphology, and VR-based cognitive-emotional signals in contrast to earlier work that concentrated on individual modalities like heart rate variability or facial morphology in isolation [10–12]. The approach is better able to capture the overall essence of Prakriti as it is presented in traditional Ayurvedic literature because of its multimodal integration. Additionally, the study presents a cognitive-affective test that is integrated into an immersive virtual reality narrative setting for the first time. With the help of this innovative setup, attention, memory, and emotional involvement may be dynamically and in real-time captured, producing neurocognitive indicators that go beyond static testing methods and supplement conventional physiological signals [ 16 – 18 ]. Lastly, the classification system, PRAK-KNN, uses a stacked ensemble learning architecture, combining interpretable meta-models (KNN and SVC) with robust base learners (Random Forest, Gradient Boosting, Extra Trees, XGBoost). High performance and local interpretability are guaranteed by this approach. The discriminative power of the chosen biomarkers is further increased using a hybrid feature selection approach that combines Random Forest importance ranking and analysis of variance, resulting in repeatable and explicable classification results. The study has three objectives. First, to ascertain if combined physiological and neuropsychological markers may be used to objectively classify constitution types. The second step is to determine which features have the biggest impact on classification performance in order to simplify the model in the future. Third, to assess if it is feasible to use VR-based challenges to identify dynamic, real-time emotional and cognitive signatures that improve classification accuracy. The study addresses long-standing concerns regarding the empirical foundation of constitution-based governments in order to achieve these goals. Does a person's constitution as described in Ayurvedic literature match their neurophysiological signature? Can wearable and immersive technology be used to regularly examine and evaluate these traits? Lastly, is it possible to make such profiling sufficiently reliable and interpretable for practical uses? Practically speaking, this work establishes the groundwork for digital diagnostics that are sensitive to the constitution and can be integrated into mobile applications, wellness centers, and remote health platforms. The suggested paradigm provides a scientifically sound link between traditional models of individualized treatment and contemporary AI-driven health systems by eliminating subjectivity and facilitating real-time, repeatable evaluation. The dataset, feature engineering procedures, signal processing techniques, and machine learning pipeline are all thoroughly explained in the parts that follow. The accuracy, F1-score, and cross-validation are then used to assess the model's performance. The discussion concludes by placing the results in the context of previous research and outlining possible directions for clinical implementation and long-term observation. 2. Methods 2.1 Subjects A total of 48 people (22 men and 26 women), aged 18 to 60 years, freely participated in the research. The participants were screened to rule out those with a history of acute or chronic neurological, cardiovascular, or mental illness. Additionally, individuals with hearing disabilities or any condition that limits the use of a VR headset were ruled out. Informed consent was obtained from all subjects. The demographic profile ensured that there was a representative sample across different Prakriti categories, which is indispensable for developing a robust classification model. 2.2 Compilation of Multimodal Dataset and Physiological-Cognitive Profiling A large-scale multimodal dataset was constructed to represent a wide range of individualised constitution-related physiological and cognitive properties. The study employed the union of neuroscience, physiological monitoring, thermal imaging, computer vision, and VR-based cognitive assessment methods. All data modalities were selected by their capacity to record dynamic human physiological and cognitive properties to create a firm diagnostic foundation for machine learning classification. 2.2.1 Electroencephalography (EEG) EEG was conducted using the Muse S-band EEG system [ 19 ], with the ability to non-invasively record basic brain activity patterns outside the clinical environment. The technology enabled real-time acquisition of brainwave frequencies in the delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), and gamma (30–100 Hz) frequency bands. These frequencies are closely associated with particular states of cognition, such as stress response, attention, memory encoding, and relaxation [ 23 ]. EEG measures were taken both when the subjects were resting and during cognitive stimulation phases to monitor the shift in engagement, cognitive load, and affective reactivity between constitution types. 2.2.2 Heart Rate Variability (HRV) Heart rate variability (HRV) was quantified using the EmWave Pro system [ 21 ] with non-invasive evaluation of the balance between the autonomic nervous system. Main temporal and spectral parameters such as very low frequency (VLF: 0.0033–0.04 Hz), low frequency (LF: 0.04–0.15 Hz), and high frequency (HF: 0.15–0.4 Hz) bands, RR intervals, and SDNN were computed. These HRV indicators are responsive to cardiac resilience, autonomic adaptability, and emotional regulation of stress [ 18 ]. The measurements were made in resting, cognitive task performance, and task recovery states in an attempt to acquire dynamic physiological readjustments. 2.2.3 Thermography Facial thermal imaging was conducted with a FLIR infrared camera, targeting dynamic temperature changes on the forehead, cheek, and nose areas. Forehead and cheek temperature changes were used as indicators of cognitive workload and emotional arousal, and nose tip temperature decreases were used to indicate acute stress or anxiety responses [22]. This contactless modality offered supplemental physiological information indicative of affective and cognitive states. 2.2.4 Facial and Lingual Feature Analysis High-resolution frontal facial images were taken under controlled lighting. Facial landmark detection and feature extraction pipelines were used automatically to measure facial morphology, symmetry, and textural features [ 23 ]. Simultaneously, tongue images were examined for colour, texture, and shape features parameters conventionally linked to metabolic and digestive health markers in constitution profiling paradigms [ 24 ]. In combination, facial and lingual biomarkers strengthened the discriminative ability of the model, especially for distinguishing between constitution types with minor physiological variations. 2.2.5 Ambulatory Assessment The gait and postural features of participants were captured by standardised walking tests [ 25 ]. Frame-by-frame movement tracking allowed the extraction of stride length, cadence, and postural sway parameters. Convolutional neural network (CNN)-based biomechanical models were used to analyse data to determine constitution-specific movement signatures [ 25 ]. Earlier theoretical accounts posit that Vata individuals have light and unsteady movements, Pitta individuals show quick and purposeful walks, and Kapha individuals demonstrate stable, rooted gaits, predictions that were tested empirically in the current study. 2.2.6 Virtual Reality (VR) based cognitive test To record cognitive-affective reactions in ecologically valid environments, participants interacted with a culture-engaging VR module created through CoSpaces Edu [ 26 ]. The VR story, based on the life of King Harishchandra, featured ethically challenging situations and emotional shifts intended to create cognitive load and emotional investment [ 27 ]. Ongoing EEG and HRV recordings were taken during VR exposure to track real-time physiological responses. After the VR experience, participants rated a multiple-choice questionnaire of memory retention and a verbal narrative recall task, allowing quantification of Shrut Grahi auditory intelligence and emotional processing. 2.2.7 Integrated Multimodal Dataset Combining EEG, HRV, thermal imaging, facial and tongue characteristics, gait analysis, and VR-based cognitive testing provided a rich, multi-dimensional dataset consisting of 214 features. This comprehensive data basis was used as the input for the following machine learning pipelines to execute robust and objective constitution classification. 2. 3 Stages of Input Variables and Data Collection To monitor physiological and cognitive profiles systematically, the experiment applied a systematic four-phase observation. Baseline tests were performed during the pre-virtual reality phase for about five minutes to examine subjects' resting cognitive and physiological function. Parameters under observation were electroencephalography (EEG) to capture baseline cerebral function, heart rate variability (HRV) to measure determination of autonomic function, thermography to capture early thermal profiling, and facial and tongue imaging to perform structural assessment. The 10- to 15-minute virtual reality part consisted of interaction with a culture-adapted VR exercise by the participants. EEG and HRV were collected in real-time during this part to measure continuous changes in cognitive and emotional responses elicited by the virtual setting. Immediately following the VR, the participants entered the cognition phase, where they completed a recall survey of the VR story. At the same time, ongoing EEG and HRV recordings monitored memory processes and cognitive load. The conclusive five-minute phase was the second stage of physiological measurements to assess recovery procedures. Electroencephalography (EEG), heart rate variability (HRV), thermography, facial and tongue imaging, and gait analysis were applied to record the dynamics of normalisation of physiology to cognitive-emotional stimulation. The constitution type of the participants (Prakriti types) was chosen through a standardised clinical evaluation at the Regional Ayurveda Research Institute (RARI) at Mandi, Himachal Pradesh, India. The evaluations were performed by an Ayurveda-qualified physician with more than five years of clinical diagnostic experience. The evaluations were done using established criteria from traditional Ayurvedic texts and employed validated Prakriti testing questionnaires from previous clinical studies. The diagnostic process was a total consultation of physical, physiological, and mental characteristics like body type, metabolic tendencies, skin, emotional responses, and intellectual styles. After the first test using questionnaires, the physician confirmed the results by clinical observation and examination to finish the constitution classification. Table 1 shows the assignment of the constitution among the 48% of participants. As seen in Table 1 , the Prakriti label Pitta-Kapha consisted of 18 participants and 37.5% of the data. Next, the Kapha, Pitta, Kapha-Vata and Vata-Pitta labels contained 10 (20.8%), 08 (16.7%), 08 (15.6%) and 04 (8.3%) of data, respectively. The labels Vata and Vata-Pitta-Kapha did not contain any labels in the collected data ( Table 1 ) below. Thus, among the 48 patients, 18 (37.5%) fulfilled a single Prakriti label; the other 30 subjects (62.5%) had mixed-type Prakritis (Pitta-Kapha, Kapha-Vata, Vata-Pitta) and were considered as separate classes in machine learning. This clinical validation approach ensured the ground-truth labels made available for model training and testing were reliable. Table 1 The Assignment of the Constitution among the 48 participants. Prakriti label Participants Share of cohort Pitta-Kapha 18 37.5% Kapha 10 20.8% Pitta 08 16.7% Kapha-Vata 08 15.6% Vata-Pitta 04 8.3% Vata 00 0% Vata-Pitta-Kapha 00 0% 2.4 Machine Learning Pipelines and Feature Engineering A streamlined machine learning (ML) pipeline was implemented to simplify data preprocessing, dimensionality reduction, model training, and performance evaluation. To make data preparation, dimensionality reduction, model training, and performance evaluation easier, a simplified machine learning (ML) pipeline was put into place. EEG, HRV, thermography, face imaging, gait analysis, and VR-based cognitive-emotional indicators were among the six modalities from which the pipeline first identified 214 characteristics after processing data from n = 48 subjects. The dataset was narrowed down to 69 high-impact characteristics for model training after feature selection using ANOVA and random forest priority ranking. To preserve the accuracy and dependability of the inferences made from the multimodal dataset, each step was built methodically. Figure 1 depicts the machine learning pipeline's complete design. Each step was constructed systematically to maintain the integrity and reliability of the conclusions drawn from the multimodal dataset. The creation of the PRAK-KNN model, an interpretable K-nearest-neighbours-based meta-classifier that combines ensemble meta-features with customized proximity-based learning, was a significant advancement in this process. 2.5 Data Preprocessing and Feature Extraction EEG activity was recorded at a 256 Hz sampling rate via the Muse S headband, a dry 4-channel electrode unit. Under the default montage of the Muse system, which approximates the global 10–20 system, electrodes were located on TP9, AF7, AF8, and TP10. Five minutes of eyes-closed recording and five minutes of eyes-open recording were included in each session to yield a stable baseline measurement. EEG signals were band-pass filtered to denoise and subsequently wavelet decomposed to eliminate artifacts [ 28 ]. Power spectral density (PSD) was also computed to give a quantitative description of signal energy distribution as a function of standard brain-wave frequency bands [ 29 ]. The EmWave Pro gadget, which measures heart rhythm patterns at a sampling rate of 125 Hz utilising infrared pulse plethysmography, was used to capture HRV data. Ten minutes were spent in a seated, resting position throughout each recording session. After spectrum analysis and artifact reduction, time-domain metrics (such as RMSSD and SDNN) and frequency-domain indices (like LF/HF ratio) were retrieved. HRV data were analysed using a fast Fourier transform (FFT) to quantify time-domain parameters such as RMSSD and SDNN, as well as frequency-domain indices such as the LF/HF ratio [ 30 ]. Thermal-imaging data were filtered to calculate mean temperature gradients for predefined facial regions of interest [ 31 ]. Gait features were obtained from video recordings by running OpenPose and examining the following movement trajectories using three-dimensional convolutional neural network (3D CNN) models in a bid to preserve spatial-temporal dynamics [32]. Facial and tongue images were localised using landmark localisation, and principal-component analysis (PCA) was utilised to reduce dimensions while protecting the most informative colour and shape feature vectors [ 33 ] in Fig. 1 below. 2 .6 Feature Selection and Standardisation A consistent-classification feature-space optimisation, a hybrid approach of feature selection, was employed. Analysis of variance (ANOVA) was used [ 34 ] to select features that showed statistically significant variation among the constitution groups. Along with this process, we approximated feature-importance values using random forests and ranked them based on their contribution to increased model performance in Gini importance [ 35 ]. This dual strategy ensured that the selected features were not only statistically robust but also pragmatically effective in the problem of classification, resulting in an understandable and well-balanced feature set. 2.7 Architectural Framework The machine-learning architecture utilised a two-level ensemble framework. Base learners utilized are Random Forest (RF) [ 35 ], Gradient Boosting (GBoost) [ 36 ], Extra Trees [ 37 ], and Extreme Gradient Boosting (XGBoost) [ 38 ], all of which were selected for their capabilities with unbalanced classes and multi-modal feature spaces, as well as for being interpretable. Meta-models like logistic regression (LR) [ 39 ], support-vector classifier (SVC) [40], and K-nearest neighbours (KNN) [ 41 ] combined base-learner predictions to promote prediction performance. Among them, PRAK-KNN was presented as a strong meta-model that improves classification customisation by utilizing neighbourhood-based learning on top of ensemble meta-features. A stacked ensemble was applied, wherein meta-features generated by base learners were used as input to the meta-classifiers [ 42 ]. Stability of the models was achieved with stratified five-fold cross-validation [43], and hyperparameter tuning was performed through a combination of grid search and random search strategies [ 44 ]. The accuracy of the classification models was confirmed by employing a blend of various statistical metrics to determine overall validation [ 45 ]. Accuracy was employed to measure over all correctness of model predictions, whereas precision calculated the actual positive predictions from all total predictions as positive. Recall verified the model's capacity to detect all pertinent instances, and the F1-score gave a balanced estimation through the combination of precision and recall [ 45 ]. All machine learning models were implemented in Python using the scikit-learn (v1.2.0), XGBoost (v1.7.4), and LightGBM (v3.3.5) libraries. OpenPose (v1.7.0), which contained a 3D CNN developed in PyTorch (v1.13), was used to perform gait analysis. MNE-Python was used to analyse the EEG data and extract features (v1.3.1). SciPy (version 1.10.1) and pandas (version 1.5.3) were used for the remaining statistical studies. 2.8 Model Deployment and Output Variables Following feature optimisation, machine learning models were used to identify the implicit relationships between multimodal physiological-cognitive signals and constitution types [ 35 – 38 ]. The models RF, GBoost, ET, and XGBoost were employed since they are established to be strong and able to handle intricate interactions among features [ 35 – 38 ]. Ensemble-based learning techniques enabled model generalisation by pooling various patterns from various modalities [ 42 ]. Because of its local decision bounds based on ensemble predictions, PRAK-KNN may continuously beat conventional models in tailored Prakriti categorization. Cross-validation strategies ensured non-bias and increased external validity of models [43]. Final validation with accuracy, precision, recall, and F1-score ensured replicability and reliability of the proposed framework [ 45 ], hence giving a scientifically verified approach towards individual constitution classification based on multimodal biometrics. 2.9 Statement of Ethics The Indian Institute of Technology Mandi's Institutional Ethics Committee gave its approval to the study protocol (Approval Number: IITM/IEC(H)/2025/VD/P2). Every experiment involving human subjects was carried out in compliance with the 1964 Declaration of Helsinki's ethical guidelines and any subsequent revisions. Before being included in the study, each subject gave their informed written permission. The study's objectives, methods, and participants' freedom to discontinue participation at any moment without facing repercussions were explained to them. 3. Results The full multimodal dataset had 214 features derived from physiological, morphological, and cognitive measurements in 48 participants. A hybrid approach merging statistical and machine learning methods was used to select stable features. A statistical test, Analysis of Variance (ANOVA), was first run to statistically test the discriminatory power of each feature for constitution types. At the same time, a Random Forest classifier was used to train the classification by assigning Gini importance scores to features ranked according to how much each contributed to model accuracy. A combined ranking was then created by combining outcomes from both ANOVA and Random Forest analysis so that the end-features chosen were both statistically relevant and predictively significant This two-stage strategy resulted in the development of three feature sets: Top 10 Features, Top 30 Features, and the Full Feature Set of all 214 features. Some of the most discriminative features were Relative Beta at TP10 during VR, Absolute Beta at TP10 during VR, and Relative Theta at TP9 post VR, which emphasised the critical role played by cognitive effort elicited through engaging VR stories. HRV-based measures like pre-HF power, during-HF power, and post-RMSSD also had the highest scoring, as per theoretical expectations of autonomic variability with constitution types. Structural and thermophysiological attributes like Post-Exposure Eye Aspect Ratio (EAR) and Absolute Alpha at AF7 pre-VR also contributed toward classification accuracy, establishing structural and thermophysiological characteristics to be salient to constitution profiling. The findings rigorously confirmed the study hypothesis that multimodal feature integration would improve classification performance. Performance metrics for base models, meta-learners, and stacked ensembles were increasingly tested as feature richness increased from 10 to 30 to all features available. 3.1 Results with 10 Features With the first evaluation using the Top 10 Features, Extra Trees posted the highest base model accuracy at 64%, followed closely by Gradient Boosting (61%) and Random Forest (60%), as seen in ( Table 2 ). Table 2 Performance of base models in Prakriti prediction using the best 10 features. Model Accuracy Precision Recall F1-score RF 0.60 0.55 0.52 0.51 GBOOST 0.61 0.66 0.58 0.57 EXTRATREE 0.64 0.62 0.55 0.55 XGBOOST 0.57 0.69 0.61 0.60 In the second stage, the PRAK-KNN model ranks the best with 65% accuracy. Among the meta-models, as shown in (Table 3 ) , the Support Vector Classifier (SVC) is followed at 63%, and the Logistic Regression is at 35%. The results indicate that good classification performance was achievable even with a minimal set of features, albeit with moderate restrictions. Table 3 Performance of meta models in Prakriti prediction using the best 10 features. Meta Model Accuracy Precision Recall F1-score LR 0.35 0.29 0.31 0.25 SVC 0.63 0.67 0.53 0.53 PRAK-KNN 0.65 0.71 0.60 0.60 3.2 Results with 30 Features Scaling up to the top 30 features resulted in a dramatic improvement in model performance. Extra Trees and XGBoost both 70%, Gradient Boosting scored 68%, and Random Forest scored 64% ( Table 4 ) . At the ensemble level, the PRAK-KNN framework outperformed others: the K-Nearest Neighbours meta-learner achieved 70%, followed closely by SVC at 69% and Logistic Regression at 58% ( Table 5 ). This indicates that the PRAK-KNN framework is particularly effective in this context, leveraging the strengths of the K-Nearest Neighbours method. Table 4 Performance of base models in Prakriti prediction using the best 30 features. Model Accuracy Precision Recall F1-score RF 0.64 0.68 0.64 0.63 GBOOST 0.68 0.78 0.71 0.70 EXTRATREE 0.71 0.66 0.64 0.62 XGBOOST 0.70 0.65 0.64 0.62 Table 5 Performance of meta models in Prakriti prediction using the best 30 features. Meta Model Accuracy Precision Recall F1-score LR 0.58 0.41 0.42 0.38 SVC 0.69 0.78 0.68 0.69 PRAK-KNN 0.70 0.74 0.65 0.66 These findings confirm that including more physiological and cognitive features, especially those of VR-based cognitive engagement, significantly improved model predictability and resilience. 3.3 Results with All Features When using the full feature set of all 214 features, the models achieved their best performances. Extra Trees obtained the highest base model accuracy at 76%, followed closely by XGBoost at 75%, Random Forest at 72%, and Gradient Boosting at 71%, as seen in ( Table 6 ). In the PRAK-KNN stack, the K-Nearest Neighbours meta-learner achieved 76% accuracy with an F1-score of 0.75, while the Support Vector Classifier recorded 75% accuracy ( Table 7 ). Table 7 highlights the comparative performance of these models, indicating that the PRAK-KNN stack has a slight edge in accuracy and F1-score. This suggests that the K-Nearest Neighbours meta-learner may be more effective in handling the specific characteristics of the dataset used in this study. The confusion matrix is shown in Fig. 2 . Table 6 Performance of base models in Prakriti prediction using all features. Model Accuracy Precision Recall F1 score RF 0.72 0.79 0.72 0.72 GBOOST 0.71 0.76 0.71 0.70 EXTRATREE 0.76 0.77 0.71 0.71 XGBOOST 0.75 0.76 0.68 0.67 Table 7 Performance of meta models in Prakriti prediction using all features. Meta Model Accuracy Precision Recall F1-score LR 0.47 0.37 0.41 0.35 SVC 0.75 0.79 0.71 0.71 PRAK-KNN 0.76 0.82 0.74 0.73 Overall, the findings reliably validated the original research hypotheses. Progressive feature augmentation resulted in related improvements in classification performance. The results illustrated that EEG signals collected during VR exposure, HRV-based autonomic indicators, facial morphology, thermal imaging, and gait dynamics each made significant contributions to constitution classification. Moreover, in the ensemble stage, PRAK-KNN employed KNN and SVC as meta-learners, performed better than less sophisticated models, and were effective in capturing the complex and holistic nature of multimodal physiological-cognitive signals. Hence, the results strongly substantiate the scientific hypothesis that a holistic, unbiased, and multimodal system can effectively characterise individual constitutions, opening doors to future applications in personalised healthcare systems. 4. Discussion This research introduced PRAK-KNN, a model that describes a machine learning framework capable of objectively profiling the health types of individuals through a blend of multimodal physiological and neurocognitive signs. Through a combination of heart rate variability (HRV), electroencephalography (EEG), facial thermal imaging, gait patterns, and cognitive-emotional reactions in immersive virtual-reality (VR) contexts, this work surpasses subjective judgments conventionally linked with constitution-based health profiling [ 46 ]. The findings add credence to the idea that combining several cognitive and physiological streams improves the accuracy and dependability of categorization. When compared to enriched feature sets, early trials using reduced feature sets produced poorer classification accuracies. But as the combined data's richness increased, so did the models' capacity for prediction. Classification accuracies of up to 76% were achieved by models built using the complete multimodal feature set, while ensemble learners like Extra Trees and XGBoost outperformed simple techniques in general. These accuracies are higher than those of earlier studies in the field, which used single modalities such as HRV or facial morphology and would generally reach accuracies between 60–65% [47, 48]. In the PRAK-KNN framework, the optimal meta-learners (KNN and SVC) achieved the 76% benchmark and produced consistent F1-scores. These results show how well the PRAK-KNN stack applies these algorithms to sustain high performance on a range of datasets. Of particular interest, the research highlights the key role of dynamic neurocognitive measures recorded during VR narrative sessions. Cognitive load and emotional-arousal features were highly discriminative, indicating that moment-to-moment cognitive reactions, as well as static physiological indicators, enhance individual profiling. This result aligns with current research in immersive cognition and affective computing, which demonstrates how VR-based cognitive tasks can distinguish between individuals based on their executive control, emotional reactivity, and attention regulation skills [ 53 , 54 ]. This is consistent with embodied-cognition theories, which assume that cognitive and physiological states are strongly interconnected [49, 50]. The findings also corroborate the general concept that constitutional health types are a multidimensional integration of stable physiological characteristics and adaptive cognitive-emotional patterns. Conventional constitution-framed models have traditionally assumed that inter-individual variability includes both innate characteristics and variable responses; our data-based conclusions lend empirical validation to this hypothesis [ 51 ]. Ensemble-learning approaches not only improved accuracy but also provided stability across varying feature sets, which points towards their applicability in real-world personalised-medicine use cases [ 52 ]. Despite the promising outcomes with PRAK-KNN, it is important to recognize a number of limitations. The emotional stimuli and story frameworks were culturally rooted, even if the VR-based cognitive exam recorded valuable information. It would be necessary to modify the VR material to accommodate various cultural and cognitive situations to generalise this framework. Furthermore, the resolution limitations of the wearable devices (Muse S and EmWave Pro) may restrict fine-grained inference even if EEG and HRV offer profound insight into autonomic and cerebral control. Subjectivity is introduced since the labelling of the constitution was determined by experts. The rigor of future validation might be increased by using proteomic or genomic ground truths. In the future, the integration of real-time wearable sensors and longitudinal monitoring platforms can transform individualised health profiling into adaptive, continuous systems. Merging the suggested framework with real-time biofeedback may also facilitate early detection of stress, disease susceptibility, and personalized wellness interventions. PRAK-KNN has potential applications in remote wellness centres, telemedicine platforms, and preventive health analytics. It can help clinicians triage patients according to constitutionally linked risk profiles. Embedding this model into mobile health apps could offer interactive, user-friendly dashboards that visualise Prakriti-linked physiology for lay users and practitioners alike. Another issue is the models' interpretability, which makes it simple for users and healthcare professionals to understand and react to model outputs. In order to improve user acceptance and confidence, future work might benefit from including explainable AI (XAI) tools like SHAP or LIME, which provide feature-level insights on predictions. PRAK-KNN incorporates physiological parameters with cognitive-emotional dynamics to equip a replicable, data-driven framework for constitution profiling, honouring unique variances while adhering to contemporary scientific standards and facilitating truly personalised therapy. 5. Conclusion This research presents convincing evidence that patient-specific health profiling can be maximally improved with the use of PRAK-KNN, a multimodal machine learning-based framework that couples physiological signals with neurocognitive markers. An integration of EEG, HRV, thermal, gait dynamic, and virtual reality (VR)-based cognition tests provides this method with a replicable, objective alternative for subjective assessment methods. The increasing improvements in classification accuracy seen as richness in multimodal data only highlight the significance of combining more than one biosignal and cognitive response. Ensemble models were much superior as compared to simple classification models in multiple cases. This demonstrates the value of employing robust model architectures in uncovering the complex relationship between physiological states and cognitive-emotional processes. The findings are also in agreement with contemporary embodied cognition theories that centre on the interdependency of mind and body functions [50]. The inclusion of real-time cognitive load and emotional arousal measures offers a better overall explanation of within-person variability, with vast potential for future personalised healthcare applications. While the present work was centred on constitution-based profiling, the method is scalable to larger areas of application, like customised mental health assessment and adaptive well-being interventions. Future research directions should be toward increasing the variability of cognitive and affective stimuli and incorporating wearable technologies in real time to enable continuous and adaptive health profiling. In conclusion, this study illustrates that PRAK-KNN effectively integrates machine learning-driven multimodal health profiling with traditional holistic approaches. This integration paves the way for enhanced precision in personalised and adaptive healthcare solutions. Declarations Ethics Approval This study was approved by the Institutional Ethics Committee of the Indian Institute of Technology Mandi (IIT Mandi), India (Approval No. IITM/IEC(H)/2025/VD/P2). All procedures involving human participants were conducted in accordance with the ethical standards of the institutional research committee and the 1964 Helsinki declaration and its later amendments. Informed written consent was obtained from all participants prior to their inclusion in the study. Data Availability The datasets generated and/or analyzed during the current study are not publicly available due to participant confidentiality constraints but are available from the corresponding author on reasonable request. Code Availability The code used for preprocessing, feature extraction, and model training is available from the corresponding author upon reasonable request. Author Contributions V.D. conceived the study, supervised the research, and contributed to manuscript writing. A.B. assisted in experimental design and guided multimodal data integration. N.B. implemented the machine learning pipeline, performed the data analysis, and led manuscript drafting. A.S., G.C., R.S., and A.C. contributed to data collection and organisation. All authors reviewed the manuscript and approved the final version for submission. Competing Interests The authors declare no competing interests. Acknowledgement We would like to acknowledge the financial support of the All India Council for Technical Education's grant to Prof. Varun Dutt (PI). Also, we would like to thank the Indian Institute of Technology Mandi for helping us with computational resources for the project. References Dash, S., Sharma, M., Mitra, P. Multimodal biometric systems: An overview. Pattern Recognition 93, 379–393 (2020). Roy, S., Bhattacharyya, D., Kim, T.H. Multimodal biometric fusion techniques: A review. Int. J. Comput. Inf. Syst. Ind. Manag. Appl. 12, 235–245 (2020). D’Angelo, S., Eickhoff, S.B. Virtual reality and neurocognitive assessment: Toward smarter models. Front. Psychol. 11, 605925 (2020). Madole, A., Joshi, A., Bharath, R.D. Virtual reality paradigms for assessment of cognitive functioning: A review. Sci. Rep. 12, 4650 (2022). Zhao, X., Zhang, L., Yang, Z. Combining multimodal physiological signals for emotion recognition using ensemble methods. IEEE Trans. Affect. Comput. 12, 1108–1121 (2021). Juyal, R.C., Mishra, M., Nautiyal, R., Prasher, B. Ayurgenomics: A new approach in personalized and preventive medicine. Sci. Cult. 78, 478–482 (2012). Joshi, R., Bhatt, N., Ghodke, Y., Patwardhan, B. Ayurvedic concept of Prakriti and its scientific validation through genetic markers. Indian J. Med. Res. 149, 35–41 (2019). Sharma, H., Chandola, H.M., Singh, G., Basisht, G. Genetic basis of Ayurveda: Ayurgenomics. Evid. Based Complement. Alternat. Med. 2015, 1–7 (2015). Shankar, D., Tiwari, R., Kumar, P. Validation of Prakriti-based Ayurvedic diagnostics using modern scientific tools. Front. Physiol. 11, 585960 (2020). Mishra, M., Prasad, R., Tiwari, A. Facial analysis for Prakriti classification using machine learning. J. Ayurveda Integr. Med. 8, 60–65 (2017). Bidve, P., Mishra, S., Annapurna, J. Enhancing Ayurvedic diagnosis using multinomial naive Bayes and K‑modes clustering. arXiv:2310.02920 (2023). HeartMath Institute. Heart rate variability and biofeedback research. HeartMath Research Library (Accessed 2024). Muse. Muse S: The sleep and meditation headband. Interaxon Inc. (Accessed 2024). Mehta, P., Singh, A., Mehra, S. Virtual reality in Ayurvedic cognitive assessment: A pilot study. J. Tradit. Complement. Med. 12, 240–251 (2022). Rotti, H., Raval, R., Anchan, S., et al. Determinants of Prakriti and its correlation with genomic variability. BMC Complement. Altern. Med. 14, 1–10 (2014). Bhushan, P., Kalpana, J., Arvind, C. Classification of human population based on HLA gene polymorphism. J. Altern. Complement. Med. 11, 349–353 (2005). Interaxon Inc. Muse S‑band EEG system User Manual (2024). Smith, J.A., et al. Cognitive correlates of EEG frequency bands: A review. Neurosci. Biobehav. Rev. 50, 194–203 (2015). HeartMath LLC. EmWave Pro Personal Stress Reliever Guide (2023). Pavlidis, I., et al. Thermal imaging for stress detection: A review. IEEE Trans. Affective Comput. 3, 102–115 (2012). Kazemi, V., Sullivan, J. One millisecond face alignment with ensemble of regression trees. Proc. CVPR, 1867–1874 (2014). Patil, R., et al. Tongue image analysis in health screening: A survey. Comput. Med. Imaging Graph. 58, 35–45 (2017). Han, J., et al. Gait recognition via CNNs: A comprehensive study. Pattern Recognit. 87, 1–15 (2019). CoSpaces Edu. Virtual Reality Storytelling Platform Documentation (2024). Chatterjee, S., et al. King Harishchandra narrative in cognitive VR tests. J. Virtual Real. Med. 5, 45–53 (2021). Sanei, S., Chambers, J.A. EEG Signal Processing. Wiley (2007). Niedermeyer, E., da Silva, F.L. Electroencephalography: Basic Principles, Clinical Applications. Lippincott (2005). Shaffer, F., Ginsberg, J.P. Heart Rate Variability Metrics and Norms. Front Public Health, 5, 258 (2017). Ioannou, S., Gallese, V., Merla, A. Thermal imaging in psychophysiology. Psychophysiology, 51(10), 951–963 (2014). Cao, Z., Simon, T., Wei, S.E., Sheikh, Y. Realtime Pose Estimation using PAF. Proc. CVPR, 1302–1310 (2017). Sagonas, C., Tzimiropoulos, G., Zafeiriou, S., Pantic, M. Facial landmark annotation. Proc. CVPR Workshops, 896–903 (2013). Field, A. Discovering Statistics Using IBM SPSS Statistics (4th ed.). Sage (2013). Breiman, L. Random Forests. Machine Learning, 45(1), 5–32 (2001). Friedman, J.H. Greedy Function Approximation. Annals of Statistics, 29(5), 1189–1232 (2001). Geurts, P., Ernst, D., Wehenkel, L. Extremely randomized trees. Machine Learning, 63(1), 3–42 (2006). Chen, T., Guestrin, C. XGBoost. Proc. ACM SIGKDD, 785–794 (2016). Hosmer, D.W., Lemeshow, S. Applied Logistic Regression (2nd ed.). Wiley (2000). Cortes, C., Vapnik, V. Support-vector networks. Machine Learning, 20(3), 273–297 (1995). Cover, T., Hart, P. Nearest neighbor pattern classification. IEEE Trans. Inf. Theory, 13(1), 21–27 (1967). Wolpert, D.H. Stacked Generalization. Neural Networks, 5(2), 241–259 (1992). Kohavi, R. Cross-validation and bootstrap. Proc. IJCAI-95, 1137–1143 (1995). Bergstra, J., Bengio, Y. Random Search for Hyper-Parameter Optimization. JMLR, 13, 281–305 (2012). Bishop, C.M. Pattern Recognition and Machine Learning. Springer (2006). Kshirsagar A., Rao P., Patwardhan B. Systems biology and Ayurveda. J. Ayurveda Integr. Med. 12(2), 187–194 (2021). Mishra M., Prasad R., Tiwari A. Facial analysis for constitution classification. J. Ayurveda Integr. Med. 8(1), 60–65 (2017). Udupa K., Sathyaprabha T.N., Thirthalli J. HRV in constitution classification. J. Ethnopharmacol. 234, 48–52 (2019). D’Angelo S., Eickhoff S.B. VR and neurocognitive assessment. Front. Psychol. 11, 605925 (2020). Wilson M. Six views of embodied cognition. Psychon Bull Rev. 9(4), 625–636 (2002). Patwardhan B., Bodeker G., Vaidya A.D.B. Ayurvedic genomics. J. Altern. Complement. Med. 21(6), 340–344 (2015). Dietterich T.G. Ensemble methods in machine learning. In MCS Proc., 1–15 (2000). Luong, T., Lécuyer, A., Martin, N., Argelaguet, F. Affective and Cognitive VR. IEEE TVCG 28, 5154–5171 (2021). Li, M., et al. Emotion-Cognition Interaction in VR. IEEE TVCG 30, 2044–2054 (2024). Bhushan, P., Kalpana, J., Arvind, C. Prakriti classification by HLA polymorphism. J. Altern. Complement. Med. 11, 349–353 (2005). 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-7044176","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":487059260,"identity":"8075b505-9f52-491b-a4e1-aec7b852c2f1","order_by":0,"name":"Navdha Bhardwaj","email":"","orcid":"","institution":"Indian Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Navdha","middleName":"","lastName":"Bhardwaj","suffix":""},{"id":487059261,"identity":"5fc93684-c98f-4231-822b-2afb9d6c8e95","order_by":1,"name":"Akash Singh","email":"","orcid":"","institution":"Indian Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Akash","middleName":"","lastName":"Singh","suffix":""},{"id":487059262,"identity":"23ca6e3d-874b-48d5-aa23-336b11fcec65","order_by":2,"name":"Ritik Sharma","email":"","orcid":"","institution":"Indian Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Ritik","middleName":"","lastName":"Sharma","suffix":""},{"id":487059263,"identity":"bf87cfc0-5cab-4548-b324-5c43ae3953d5","order_by":3,"name":"Gitanshu Chaudhary","email":"","orcid":"","institution":"Indian Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Gitanshu","middleName":"","lastName":"Chaudhary","suffix":""},{"id":487059264,"identity":"d09a7713-9d2d-42d4-bc28-4755619abce7","order_by":4,"name":"Anubha Chandel","email":"","orcid":"","institution":"Regional Ayurveda Centre","correspondingAuthor":false,"prefix":"","firstName":"Anubha","middleName":"","lastName":"Chandel","suffix":""},{"id":487059265,"identity":"f1102473-c228-4e4d-9914-0baed7069d45","order_by":5,"name":"Arnav Bhavsar","email":"","orcid":"","institution":"Indian Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Arnav","middleName":"","lastName":"Bhavsar","suffix":""},{"id":487059266,"identity":"badaf1a1-524b-4058-b6d2-5d038f0098d6","order_by":6,"name":"Varun Dutt","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzUlEQVRIiWNgGAWjYNCCCgkeMM1jACIZG4jQckaCh4c0LYxtQNVgLcQ4yeD42WfShfMsZOzFDj978KaAQZ6/gbntAV4tZ9LNpGduAzpMOs3ccI4Bg+GMA4ztBvi0SDaksUnzgrUkmEkD/cK4AehOCbxa+p8BtcwBaUn/BtJiT1ALvwTIlgaQlhywLYlEaHnGbM1zDKjldk6Z5BwDieQZhwloYeNPY7zNU1Nnzz47fZvEmz82tv3t7c/wakEHQMXMpKgfBaNgFIyCUYAVAAB24DTBugWyWQAAAABJRU5ErkJggg==","orcid":"","institution":"Indian Institute of Technology","correspondingAuthor":true,"prefix":"","firstName":"Varun","middleName":"","lastName":"Dutt","suffix":""}],"badges":[],"createdAt":"2025-07-04 07:38:29","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7044176/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7044176/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87373366,"identity":"93a7d235-8b80-4311-9b23-f35a17e27b74","added_by":"auto","created_at":"2025-07-23 07:29:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":134252,"visible":true,"origin":"","legend":"\u003cp\u003eDiagrammatic representation of the Prakriti categorisation machine learning pipeline using PRAK-KNN. The pipeline comprises the following steps: (i) loading multimodal data from EEG, HRV, thermography, gait, facial, and cognitive VR tasks; (ii) preprocessing signals and extracting features specific to each modality; (iii) selecting features using ANOVA and Gini-based ranking; (iv) standardization and dimensionality reduction; (v) training base learners (RF, GBoost, ET, XGBoost) on selected features; (vi) generating meta-features from base learner outputs; and (vii) using PRAK-KNN, a novel meta-classifier that uses K-nearest neighbors on ensemble-level predictions. Cross-validation and statistical measures (accuracy, precision, recall, and F1-score) are used to assess the model's performance to guarantee its robustness and generalizability.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7044176/v1/d882ef923ee655dc7451c7bb.png"},{"id":87373365,"identity":"3394136a-d279-4273-9336-9ba868d32171","added_by":"auto","created_at":"2025-07-23 07:29:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":28262,"visible":true,"origin":"","legend":"\u003cp\u003eillustrates the performance of the classification model across five constitution classes. The model achieved high accuracy for \"pitta-kapha\", \"kapha-vatta\", and \"vatta-pitta\" (9/10), with minimal misclassifications. However, some confusion exists between similar classes, such as \"kapha\" misclassified as \"pitta-kapha\" and \"kapha-vatta.\" Overall, the model demonstrates strong predictive ability, with most true labels aligning closely with predicted classes.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7044176/v1/012b0f47fa32871dce29ca1e.png"},{"id":88238374,"identity":"21960d1b-118d-4365-871b-b85770091a60","added_by":"auto","created_at":"2025-08-04 10:47:18","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1120390,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7044176/v1/2c4d9115-4b14-4d32-bc6e-5e902b49218f.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning Framework for Individualised Health Profiling Using Multimodal Physiological and Neurocognitive Biomarkers","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eWith the goal of matching therapeutic approaches with unique physiological, psychological, and behavioral characteristics, personalized health profiling has become a crucial component of precision medicine [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Constitution-based systems, such as Ayurveda's Prakriti model, have long offered systematic approaches to health classification, among the numerous ancient frameworks that suggest such individualized characterization. Based on innate characteristics such as temperament, bodily shape, and behavioral tendencies, Prakriti categorizes people into three constitution types: Vata, Pitta, and Kapha [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. These categories affect not just health status but also treatment responses, disease susceptibility, and lifestyle management advice.\u003c/p\u003e\u003cp\u003eHowever, diagnosis has historically depended on non-standardized physical observation and verbal interpretation, and Prakriti evaluations have historically depended on the subjective judgment of Ayurvedic practitioners. The incorporation of constitution-based systems into contemporary healthcare, where data-driven validation and scientific repeatability are necessary for clinical applicability, has been hampered by this dependence on subjective techniques [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThere is renewed interest in employing objective measurements to validate traditional health systems as a result of recent technical advancements in biomedical signal processing, wearable sensors, and machine learning [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Using heart rate variability (HRV), researchers have tried to link Prakriti types to different autonomic profiles. They have found that Kapha types have parasympathetic dominance and Pitta types frequently demonstrate more sympathetic activity [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Along with facial morphological features examined using machine learning techniques, electroencephalography (EEG) and facial thermal imaging have also demonstrated potential in identifying patterns unique to a person's composition [11,12]. Additionally, Prakriti types may correlate to genetic polymorphisms associated with immunity and metabolism, according to genomic studies in Ayurgenomics [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Despite their significance, these findings lack the coherence necessary for a reliable and clinically scalable diagnostic tool because they are still fragmented and modality-specific.\u003c/p\u003e\u003cp\u003eMore significantly, cognitive and affective aspects that are essential to constitution typologies have been overlooked in traditional evaluations. According to Ayurvedic scriptures, Pitta types are deliberate and concentrated, Kapha types are grounded yet sluggish to change, and Vata types are creative and reactive [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. These psychological profiles suggest that constitution types are neurocognitive as well as physical factors. However, the existing research lacks empirical attempts to measure these cognitive-affective markers, especially in ecologically realistic, real-time contexts [15,16]. Cognitive profiling provides a technique to close this crucial gap, particularly through immersive and interactive paradigms.\u003c/p\u003e\u003cp\u003eBy presenting a multimodal machine learning system called PRAK-KNN (Prakriti-aware k-Nearest Neighbour), which integrates physiological and neurocognitive data to objectively define constitution kinds, this study overcomes these constraints. Six major data streams are included in the pipeline: gait biometrics, facial morphology, facial thermal imaging, HRV, EEG, and auditory cognition in VR-based narrative activities [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Beyond single-modality techniques, this integrated architecture offers a scalable, reproducible approach to constitution profiling that supports the objectives of precision medicine.\u003c/p\u003e\u003cp\u003eThis study's uniqueness comes from its thorough methodological integration of several constitution profile dimensions. This study creates a unified machine learning pipeline that concurrently integrates six biometric and behavioral modalities: HRV, EEG, thermal imaging, gait analysis, facial morphology, and VR-based cognitive-emotional signals in contrast to earlier work that concentrated on individual modalities like heart rate variability or facial morphology in isolation [10\u0026ndash;12]. The approach is better able to capture the overall essence of Prakriti as it is presented in traditional Ayurvedic literature because of its multimodal integration. Additionally, the study presents a cognitive-affective test that is integrated into an immersive virtual reality narrative setting for the first time.\u003c/p\u003e\u003cp\u003eWith the help of this innovative setup, attention, memory, and emotional involvement may be dynamically and in real-time captured, producing neurocognitive indicators that go beyond static testing methods and supplement conventional physiological signals [\u003cspan additionalcitationids=\"CR17\" citationid=\"CR15\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Lastly, the classification system, PRAK-KNN, uses a stacked ensemble learning architecture, combining interpretable meta-models (KNN and SVC) with robust base learners (Random Forest, Gradient Boosting, Extra Trees, XGBoost). High performance and local interpretability are guaranteed by this approach. The discriminative power of the chosen biomarkers is further increased using a hybrid feature selection approach that combines Random Forest importance ranking and analysis of variance, resulting in repeatable and explicable classification results.\u003c/p\u003e\u003cp\u003eThe study has three objectives. First, to ascertain if combined physiological and neuropsychological markers may be used to objectively classify constitution types. The second step is to determine which features have the biggest impact on classification performance in order to simplify the model in the future. Third, to assess if it is feasible to use VR-based challenges to identify dynamic, real-time emotional and cognitive signatures that improve classification accuracy.\u003c/p\u003e\u003cp\u003eThe study addresses long-standing concerns regarding the empirical foundation of constitution-based governments in order to achieve these goals. Does a person's constitution as described in Ayurvedic literature match their neurophysiological signature? Can wearable and immersive technology be used to regularly examine and evaluate these traits? Lastly, is it possible to make such profiling sufficiently reliable and interpretable for practical uses?\u003c/p\u003e\u003cp\u003ePractically speaking, this work establishes the groundwork for digital diagnostics that are sensitive to the constitution and can be integrated into mobile applications, wellness centers, and remote health platforms. The suggested paradigm provides a scientifically sound link between traditional models of individualized treatment and contemporary AI-driven health systems by eliminating subjectivity and facilitating real-time, repeatable evaluation.\u003c/p\u003e\u003cp\u003eThe dataset, feature engineering procedures, signal processing techniques, and machine learning pipeline are all thoroughly explained in the parts that follow. The accuracy, F1-score, and cross-validation are then used to assess the model's performance. The discussion concludes by placing the results in the context of previous research and outlining possible directions for clinical implementation and long-term observation.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Subjects\u003c/h2\u003e\u003cp\u003eA total of 48 people (22 men and 26 women), aged 18 to 60 years, freely participated in the research. The participants were screened to rule out those with a history of acute or chronic neurological, cardiovascular, or mental illness. Additionally, individuals with hearing disabilities or any condition that limits the use of a VR headset were ruled out. Informed consent was obtained from all subjects. The demographic profile ensured that there was a representative sample across different Prakriti categories, which is indispensable for developing a robust classification model.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Compilation of Multimodal Dataset and Physiological-Cognitive Profiling\u003c/h2\u003e\u003cp\u003eA large-scale multimodal dataset was constructed to represent a wide range of individualised constitution-related physiological and cognitive properties. The study employed the union of neuroscience, physiological monitoring, thermal imaging, computer vision, and VR-based cognitive assessment methods. All data modalities were selected by their capacity to record dynamic human physiological and cognitive properties to create a firm diagnostic foundation for machine learning classification.\u003c/p\u003e\u003cdiv id=\"Sec5\" class=\"Section3\"\u003e\u003ch2\u003e2.2.1 Electroencephalography (EEG)\u003c/h2\u003e\u003cp\u003eEEG was conducted using the Muse S-band EEG system [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e19\u003c/span\u003e], with the ability to non-invasively record basic brain activity patterns outside the clinical environment. The technology enabled real-time acquisition of brainwave frequencies in the delta (0.5\u0026ndash;4 Hz), theta (4\u0026ndash;8 Hz), alpha (8\u0026ndash;13 Hz), beta (13\u0026ndash;30 Hz), and gamma (30\u0026ndash;100 Hz) frequency bands. These frequencies are closely associated with particular states of cognition, such as stress response, attention, memory encoding, and relaxation [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. EEG measures were taken both when the subjects were resting and during cognitive stimulation phases to monitor the shift in engagement, cognitive load, and affective reactivity between constitution types.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e2.2.2 Heart Rate Variability (HRV)\u003c/h2\u003e\u003cp\u003eHeart rate variability (HRV) was quantified using the EmWave Pro system [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] with non-invasive evaluation of the balance between the autonomic nervous system. Main temporal and spectral parameters such as very low frequency (VLF: 0.0033\u0026ndash;0.04 Hz), low frequency (LF: 0.04\u0026ndash;0.15 Hz), and high frequency (HF: 0.15\u0026ndash;0.4 Hz) bands, RR intervals, and SDNN were computed. These HRV indicators are responsive to cardiac resilience, autonomic adaptability, and emotional regulation of stress [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The measurements were made in resting, cognitive task performance, and task recovery states in an attempt to acquire dynamic physiological readjustments.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\u003ch2\u003e2.2.3 Thermography\u003c/h2\u003e\u003cp\u003eFacial thermal imaging was conducted with a FLIR infrared camera, targeting dynamic temperature changes on the forehead, cheek, and nose areas. Forehead and cheek temperature changes were used as indicators of cognitive workload and emotional arousal, and nose tip temperature decreases were used to indicate acute stress or anxiety responses [22]. This contactless modality offered supplemental physiological information indicative of affective and cognitive states.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\u003ch2\u003e2.2.4 Facial and Lingual Feature Analysis\u003c/h2\u003e\u003cp\u003eHigh-resolution frontal facial images were taken under controlled lighting. Facial landmark detection and feature extraction pipelines were used automatically to measure facial morphology, symmetry, and textural features [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Simultaneously, tongue images were examined for colour, texture, and shape features parameters conventionally linked to metabolic and digestive health markers in constitution profiling paradigms [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. In combination, facial and lingual biomarkers strengthened the discriminative ability of the model, especially for distinguishing between constitution types with minor physiological variations.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003e2.2.5 Ambulatory Assessment\u003c/h2\u003e\u003cp\u003eThe gait and postural features of participants were captured by standardised walking tests [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Frame-by-frame movement tracking allowed the extraction of stride length, cadence, and postural sway parameters. Convolutional neural network (CNN)-based biomechanical models were used to analyse data to determine constitution-specific movement signatures [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Earlier theoretical accounts posit that Vata individuals have light and unsteady movements, Pitta individuals show quick and purposeful walks, and Kapha individuals demonstrate stable, rooted gaits, predictions that were tested empirically in the current study.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\u003ch2\u003e2.2.6 Virtual Reality (VR) based cognitive test\u003c/h2\u003e\u003cp\u003eTo record cognitive-affective reactions in ecologically valid environments, participants interacted with a culture-engaging VR module created through CoSpaces Edu [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. The VR story, based on the life of King Harishchandra, featured ethically challenging situations and emotional shifts intended to create cognitive load and emotional investment [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Ongoing EEG and HRV recordings were taken during VR exposure to track real-time physiological responses. After the VR experience, participants rated a multiple-choice questionnaire of memory retention and a verbal narrative recall task, allowing quantification of Shrut Grahi auditory intelligence and emotional processing.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\u003ch2\u003e2.2.7 Integrated Multimodal Dataset\u003c/h2\u003e\u003cp\u003eCombining EEG, HRV, thermal imaging, facial and tongue characteristics, gait analysis, and VR-based cognitive testing provided a rich, multi-dimensional dataset consisting of 214 features. This comprehensive data basis was used as the input for the following machine learning pipelines to execute robust and objective constitution classification.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\n\u003ch3\u003e2. 3 Stages of Input Variables and Data Collection\u003c/h3\u003e\n\u003cp\u003eTo monitor physiological and cognitive profiles systematically, the experiment applied a systematic four-phase observation. Baseline tests were performed during the pre-virtual reality phase for about five minutes to examine subjects' resting cognitive and physiological function. Parameters under observation were electroencephalography (EEG) to capture baseline cerebral function, heart rate variability (HRV) to measure determination of autonomic function, thermography to capture early thermal profiling, and facial and tongue imaging to perform structural assessment.\u003c/p\u003e\u003cp\u003eThe 10- to 15-minute virtual reality part consisted of interaction with a culture-adapted VR exercise by the participants. EEG and HRV were collected in real-time during this part to measure continuous changes in cognitive and emotional responses elicited by the virtual setting. Immediately following the VR, the participants entered the cognition phase, where they completed a recall survey of the VR story. At the same time, ongoing EEG and HRV recordings monitored memory processes and cognitive load. The conclusive five-minute phase was the second stage of physiological measurements to assess recovery procedures. Electroencephalography (EEG), heart rate variability (HRV), thermography, facial and tongue imaging, and gait analysis were applied to record the dynamics of normalisation of physiology to cognitive-emotional stimulation.\u003c/p\u003e\u003cp\u003e The constitution type of the participants (Prakriti types) was chosen through a standardised clinical evaluation at the Regional Ayurveda Research Institute (RARI) at Mandi, Himachal Pradesh, India. The evaluations were performed by an Ayurveda-qualified physician with more than five years of clinical diagnostic experience. The evaluations were done using established criteria from traditional Ayurvedic texts and employed validated Prakriti testing questionnaires from previous clinical studies.\u003c/p\u003e\u003cp\u003eThe diagnostic process was a total consultation of physical, physiological, and mental characteristics like body type, metabolic tendencies, skin, emotional responses, and intellectual styles. After the first test using questionnaires, the physician confirmed the results by clinical observation and examination to finish the constitution classification. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the assignment of the constitution among the 48% of participants. As seen in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the Prakriti label Pitta-Kapha consisted of 18 participants and 37.5% of the data. Next, the Kapha, Pitta, Kapha-Vata and Vata-Pitta labels contained 10 (20.8%), 08 (16.7%), 08 (15.6%) and 04 (8.3%) of data, respectively. The labels Vata and Vata-Pitta-Kapha did not contain any labels in the collected data \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e below. Thus, among the 48 patients, 18 (37.5%) fulfilled a single Prakriti label; the other 30 subjects (62.5%) had mixed-type Prakritis (Pitta-Kapha, Kapha-Vata, Vata-Pitta) and were considered as separate classes in machine learning. This clinical validation approach ensured the ground-truth labels made available for model training and testing were reliable.\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\u003eThe Assignment of the Constitution among the 48 participants.\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=\"char\" char=\".\" 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\u003ePrakriti label\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eParticipants\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eShare of cohort\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePitta-Kapha\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e37.5%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eKapha\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e20.8%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePitta\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e08\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e16.7%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eKapha-Vata\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e08\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15.6%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVata-Pitta\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e04\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e8.3%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVata\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVata-Pitta-Kapha\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Machine Learning Pipelines and Feature Engineering\u003c/h2\u003e\u003cp\u003eA streamlined machine learning (ML) pipeline was implemented to simplify data preprocessing, dimensionality reduction, model training, and performance evaluation. To make data preparation, dimensionality reduction, model training, and performance evaluation easier, a simplified machine learning (ML) pipeline was put into place. EEG, HRV, thermography, face imaging, gait analysis, and VR-based cognitive-emotional indicators were among the six modalities from which the pipeline first identified 214 characteristics after processing data from n\u0026thinsp;=\u0026thinsp;48 subjects. The dataset was narrowed down to 69 high-impact characteristics for model training after feature selection using ANOVA and random forest priority ranking. To preserve the accuracy and dependability of the inferences made from the multimodal dataset, each step was built methodically. Figure\u0026nbsp;1 depicts the machine learning pipeline's complete design. Each step was constructed systematically to maintain the integrity and reliability of the conclusions drawn from the multimodal dataset. The creation of the PRAK-KNN model, an interpretable K-nearest-neighbours-based meta-classifier that combines ensemble meta-features with customized proximity-based learning, was a significant advancement in this process.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Data Preprocessing and Feature Extraction\u003c/h2\u003e\u003cp\u003eEEG activity was recorded at a 256 Hz sampling rate via the Muse S headband, a dry 4-channel electrode unit. Under the default montage of the Muse system, which approximates the global 10\u0026ndash;20 system, electrodes were located on TP9, AF7, AF8, and TP10. Five minutes of eyes-closed recording and five minutes of eyes-open recording were included in each session to yield a stable baseline measurement.\u003c/p\u003e\u003cp\u003eEEG signals were band-pass filtered to denoise and subsequently wavelet decomposed to eliminate artifacts [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Power spectral density (PSD) was also computed to give a quantitative description of signal energy distribution as a function of standard brain-wave frequency bands [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe EmWave Pro gadget, which measures heart rhythm patterns at a sampling rate of 125 Hz utilising infrared pulse plethysmography, was used to capture HRV data. Ten minutes were spent in a seated, resting position throughout each recording session. After spectrum analysis and artifact reduction, time-domain metrics (such as RMSSD and SDNN) and frequency-domain indices (like LF/HF ratio) were retrieved. HRV data were analysed using a fast Fourier transform (FFT) to quantify time-domain parameters such as RMSSD and SDNN, as well as frequency-domain indices such as the LF/HF ratio [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThermal-imaging data were filtered to calculate mean temperature gradients for predefined facial regions of interest [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Gait features were obtained from video recordings by running OpenPose and examining the following movement trajectories using three-dimensional convolutional neural network (3D CNN) models in a bid to preserve spatial-temporal dynamics [32]. Facial and tongue images were localised using landmark localisation, and principal-component analysis (PCA) was utilised to reduce dimensions while protecting the most informative colour and shape feature vectors [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e33\u003c/span\u003e] in \u003cb\u003eFig.\u0026nbsp;1\u003c/b\u003e below.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e2\u003cb\u003e.6 Feature Selection and Standardisation\u003c/b\u003e\u003c/h2\u003e\u003cp\u003eA consistent-classification feature-space optimisation, a hybrid approach of feature selection, was employed. Analysis of variance (ANOVA) was used [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e34\u003c/span\u003e] to select features that showed statistically significant variation among the constitution groups. Along with this process, we approximated feature-importance values using random forests and ranked them based on their contribution to increased model performance in Gini importance [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. This dual strategy ensured that the selected features were not only statistically robust but also pragmatically effective in the problem of classification, resulting in an understandable and well-balanced feature set.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003e2.7 Architectural Framework\u003c/h2\u003e\u003cp\u003eThe machine-learning architecture utilised a two-level ensemble framework. Base learners utilized are Random Forest (RF) [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], Gradient Boosting (GBoost) [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e], Extra Trees [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], and Extreme Gradient Boosting (XGBoost) [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], all of which were selected for their capabilities with unbalanced classes and multi-modal feature spaces, as well as for being interpretable. Meta-models like logistic regression (LR) [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], support-vector classifier (SVC) [40], and K-nearest neighbours (KNN) [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e41\u003c/span\u003e] combined base-learner predictions to promote prediction performance. Among them, PRAK-KNN was presented as a strong meta-model that improves classification customisation by utilizing neighbourhood-based learning on top of ensemble meta-features. A stacked ensemble was applied, wherein meta-features generated by base learners were used as input to the meta-classifiers [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Stability of the models was achieved with stratified five-fold cross-validation [43], and hyperparameter tuning was performed through a combination of grid search and random search strategies [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe accuracy of the classification models was confirmed by employing a blend of various statistical metrics to determine overall validation [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. Accuracy was employed to measure over all correctness of model predictions, whereas precision calculated the actual positive predictions from all total predictions as positive. Recall verified the model's capacity to detect all pertinent instances, and the F1-score gave a balanced estimation through the combination of precision and recall [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAll machine learning models were implemented in Python using the scikit-learn (v1.2.0), XGBoost (v1.7.4), and LightGBM (v3.3.5) libraries. OpenPose (v1.7.0), which contained a 3D CNN developed in PyTorch (v1.13), was used to perform gait analysis. MNE-Python was used to analyse the EEG data and extract features (v1.3.1). SciPy (version 1.10.1) and pandas (version 1.5.3) were used for the remaining statistical studies.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e2.8 Model Deployment and Output Variables\u003c/h2\u003e\u003cp\u003eFollowing feature optimisation, machine learning models were used to identify the implicit relationships between multimodal physiological-cognitive signals and constitution types [\u003cspan additionalcitationids=\"CR36 CR37\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe models RF, GBoost, ET, and XGBoost were employed since they are established to be strong and able to handle intricate interactions among features [\u003cspan additionalcitationids=\"CR36 CR37\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Ensemble-based learning techniques enabled model generalisation by pooling various patterns from various modalities [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Because of its local decision bounds based on ensemble predictions, PRAK-KNN may continuously beat conventional models in tailored Prakriti categorization. Cross-validation strategies ensured non-bias and increased external validity of models [43]. Final validation with accuracy, precision, recall, and F1-score ensured replicability and reliability of the proposed framework [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e], hence giving a scientifically verified approach towards individual constitution classification based on multimodal biometrics.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003e2.9 Statement of Ethics\u003c/h2\u003e\u003cp\u003e The Indian Institute of Technology Mandi's Institutional Ethics Committee gave its approval to the study protocol (Approval Number: IITM/IEC(H)/2025/VD/P2). Every experiment involving human subjects was carried out in compliance with the 1964 Declaration of Helsinki's ethical guidelines and any subsequent revisions. Before being included in the study, each subject gave their informed written permission. The study's objectives, methods, and participants' freedom to discontinue participation at any moment without facing repercussions were explained to them.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eThe full multimodal dataset had 214 features derived from physiological, morphological, and cognitive measurements in 48 participants. A hybrid approach merging statistical and machine learning methods was used to select stable features. A statistical test, Analysis of Variance (ANOVA), was first run to statistically test the discriminatory power of each feature for constitution types. At the same time, a Random Forest classifier was used to train the classification by assigning Gini importance scores to features ranked according to how much each contributed to model accuracy. A combined ranking was then created by combining outcomes from both ANOVA and Random Forest analysis so that the end-features chosen were both statistically relevant and predictively significant\u003c/p\u003e\u003cp\u003eThis two-stage strategy resulted in the development of three feature sets: Top 10 Features, Top 30 Features, and the Full Feature Set of all 214 features. Some of the most discriminative features were Relative Beta at TP10 during VR, Absolute Beta at TP10 during VR, and Relative Theta at TP9 post VR, which emphasised the critical role played by cognitive effort elicited through engaging VR stories. HRV-based measures like pre-HF power, during-HF power, and post-RMSSD also had the highest scoring, as per theoretical expectations of autonomic variability with constitution types. Structural and thermophysiological attributes like Post-Exposure Eye Aspect Ratio (EAR) and Absolute Alpha at AF7 pre-VR also contributed toward classification accuracy, establishing structural and thermophysiological characteristics to be salient to constitution profiling.\u003c/p\u003e\u003cp\u003eThe findings rigorously confirmed the study hypothesis that multimodal feature integration would improve classification performance. Performance metrics for base models, meta-learners, and stacked ensembles were increasingly tested as feature richness increased from 10 to 30 to all features available.\u003c/p\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Results with 10 Features\u003c/h2\u003e\u003cp\u003eWith the first evaluation using the Top 10 Features, Extra Trees posted the highest base model accuracy at 64%, followed closely by Gradient Boosting (61%) and Random Forest (60%), as seen in \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u003cb\u003e).\u003c/b\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\u003ePerformance of base models in Prakriti prediction using the best 10 features.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\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\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\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\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\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.60\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.55\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.52\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.51\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGBOOST\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.57\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEXTRATREE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.62\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.55\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.55\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBOOST\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.57\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.69\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.60\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\u003eIn the second stage, the PRAK-KNN model ranks the best with 65% accuracy. Among the meta-models, as shown in (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e, the Support Vector Classifier (SVC) is followed at 63%, and the Logistic Regression is at 35%. The results indicate that good classification performance was achievable even with a minimal set of features, albeit with moderate restrictions.\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\u003ePerformance of meta models in Prakriti prediction using the best 10 features.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMeta Model\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\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\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\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.35\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.31\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.25\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.63\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.53\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.53\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePRAK-KNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.60\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.60\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\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Results with 30 Features\u003c/h2\u003e\u003cp\u003eScaling up to the top 30 features resulted in a dramatic improvement in model performance. Extra Trees and XGBoost both 70%, Gradient Boosting scored 68%, and Random Forest scored 64% \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e. At the ensemble level, the PRAK-KNN framework outperformed others: the K-Nearest Neighbours meta-learner achieved 70%, followed closely by SVC at 69% and Logistic Regression at 58% \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e This indicates that the PRAK-KNN framework is particularly effective in this context, leveraging the strengths of the K-Nearest Neighbours method.\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\u003ePerformance of base models in Prakriti prediction using the best 30 features.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\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\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\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\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\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.63\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGBOOST\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.78\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEXTRATREE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.62\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBOOST\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.62\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\u003ePerformance of meta models in Prakriti prediction using the best 30 features.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMeta Model\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\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\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\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.41\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.42\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.38\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.69\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.78\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.69\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePRAK-KNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.74\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.66\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\u003eThese findings confirm that including more physiological and cognitive features, especially those of VR-based cognitive engagement, significantly improved model predictability and resilience.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Results with All Features\u003c/h2\u003e\u003cp\u003eWhen using the full feature set of all 214 features, the models achieved their best performances. Extra Trees obtained the highest base model accuracy at 76%, followed closely by XGBoost at 75%, Random Forest at 72%, and Gradient Boosting at 71%, as seen in \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e In the PRAK-KNN stack, the K-Nearest Neighbours meta-learner achieved 76% accuracy with an F1-score of 0.75, while the Support Vector Classifier recorded 75% accuracy \u003cb\u003e(\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e highlights the comparative performance of these models, indicating that the PRAK-KNN stack has a slight edge in accuracy and F1-score. This suggests that the K-Nearest Neighbours meta-learner may be more effective in handling the specific characteristics of the dataset used in this study. The confusion matrix is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\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\u003ePerformance of base models in Prakriti prediction using all features.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\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\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\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\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\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.72\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.72\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.72\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGBOOST\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEXTRATREE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.77\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBOOST\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.67\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\u003ePerformance of meta models in Prakriti prediction using all features.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMeta Model\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\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\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\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.47\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.37\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.41\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.35\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePRAK-KNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.74\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.73\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\u003eOverall, the findings reliably validated the original research hypotheses. Progressive feature augmentation resulted in related improvements in classification performance. The results illustrated that EEG signals collected during VR exposure, HRV-based autonomic indicators, facial morphology, thermal imaging, and gait dynamics each made significant contributions to constitution classification. Moreover, in the ensemble stage, PRAK-KNN employed KNN and SVC as meta-learners, performed better than less sophisticated models, and were effective in capturing the complex and holistic nature of multimodal physiological-cognitive signals.\u003c/p\u003e\u003cp\u003eHence, the results strongly substantiate the scientific hypothesis that a holistic, unbiased, and multimodal system can effectively characterise individual constitutions, opening doors to future applications in personalised healthcare systems.\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis research introduced PRAK-KNN, a model that describes a machine learning framework capable of objectively profiling the health types of individuals through a blend of multimodal physiological and neurocognitive signs. Through a combination of heart rate variability (HRV), electroencephalography (EEG), facial thermal imaging, gait patterns, and cognitive-emotional reactions in immersive virtual-reality (VR) contexts, this work surpasses subjective judgments conventionally linked with constitution-based health profiling [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe findings add credence to the idea that combining several cognitive and physiological streams improves the accuracy and dependability of categorization. When compared to enriched feature sets, early trials using reduced feature sets produced poorer classification accuracies. But as the combined data's richness increased, so did the models' capacity for prediction. Classification accuracies of up to 76% were achieved by models built using the complete multimodal feature set, while ensemble learners like Extra Trees and XGBoost outperformed simple techniques in general. These accuracies are higher than those of earlier studies in the field, which used single modalities such as HRV or facial morphology and would generally reach accuracies between 60\u0026ndash;65% [47, 48]. In the PRAK-KNN framework, the optimal meta-learners (KNN and SVC) achieved the 76% benchmark and produced consistent F1-scores. These results show how well the PRAK-KNN stack applies these algorithms to sustain high performance on a range of datasets.\u003c/p\u003e\u003cp\u003eOf particular interest, the research highlights the key role of dynamic neurocognitive measures recorded during VR narrative sessions. Cognitive load and emotional-arousal features were highly discriminative, indicating that moment-to-moment cognitive reactions, as well as static physiological indicators, enhance individual profiling.\u003c/p\u003e\u003cp\u003eThis result aligns with current research in immersive cognition and affective computing, which demonstrates how VR-based cognitive tasks can distinguish between individuals based on their executive control, emotional reactivity, and attention regulation skills [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e]. This is consistent with embodied-cognition theories, which assume that cognitive and physiological states are strongly interconnected [49, 50].\u003c/p\u003e\u003cp\u003eThe findings also corroborate the general concept that constitutional health types are a multidimensional integration of stable physiological characteristics and adaptive cognitive-emotional patterns. Conventional constitution-framed models have traditionally assumed that inter-individual variability includes both innate characteristics and variable responses; our data-based conclusions lend empirical validation to this hypothesis [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e51\u003c/span\u003e]. Ensemble-learning approaches not only improved accuracy but also provided stability across varying feature sets, which points towards their applicability in real-world personalised-medicine use cases [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e52\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eDespite the promising outcomes with PRAK-KNN, it is important to recognize a number of limitations. The emotional stimuli and story frameworks were culturally rooted, even if the VR-based cognitive exam recorded valuable information. It would be necessary to modify the VR material to accommodate various cultural and cognitive situations to generalise this framework. Furthermore, the resolution limitations of the wearable devices (Muse S and EmWave Pro) may restrict fine-grained inference even if EEG and HRV offer profound insight into autonomic and cerebral control. Subjectivity is introduced since the labelling of the constitution was determined by experts. The rigor of future validation might be increased by using proteomic or genomic ground truths.\u003c/p\u003e\u003cp\u003eIn the future, the integration of real-time wearable sensors and longitudinal monitoring platforms can transform individualised health profiling into adaptive, continuous systems. Merging the suggested framework with real-time biofeedback may also facilitate early detection of stress, disease susceptibility, and personalized wellness interventions. PRAK-KNN has potential applications in remote wellness centres, telemedicine platforms, and preventive health analytics. It can help clinicians triage patients according to constitutionally linked risk profiles. Embedding this model into mobile health apps could offer interactive, user-friendly dashboards that visualise Prakriti-linked physiology for lay users and practitioners alike.\u003c/p\u003e\u003cp\u003eAnother issue is the models' interpretability, which makes it simple for users and healthcare professionals to understand and react to model outputs. In order to improve user acceptance and confidence, future work might benefit from including explainable AI (XAI) tools like SHAP or LIME, which provide feature-level insights on predictions.\u003c/p\u003e\u003cp\u003ePRAK-KNN incorporates physiological parameters with cognitive-emotional dynamics to equip a replicable, data-driven framework for constitution profiling, honouring unique variances while adhering to contemporary scientific standards and facilitating truly personalised therapy.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis research presents convincing evidence that patient-specific health profiling can be maximally improved with the use of PRAK-KNN, a multimodal machine learning-based framework that couples physiological signals with neurocognitive markers. An integration of EEG, HRV, thermal, gait dynamic, and virtual reality (VR)-based cognition tests provides this method with a replicable, objective alternative for subjective assessment methods.\u003c/p\u003e\n\u003cp\u003eThe increasing improvements in classification accuracy seen as richness in multimodal data only highlight the significance of combining more than one biosignal and cognitive response. Ensemble models were much superior as compared to simple classification models in multiple cases. This demonstrates the value of employing robust model architectures in uncovering the complex relationship between physiological states and cognitive-emotional processes.\u003c/p\u003e\n\u003cp\u003eThe findings are also in agreement with contemporary embodied cognition theories that centre on the interdependency of mind and body functions [50]. The inclusion of real-time cognitive load and emotional arousal measures offers a better overall explanation of within-person variability, with vast potential for future personalised healthcare applications.\u003c/p\u003e\n\u003cp\u003eWhile the present work was centred on constitution-based profiling, the method is scalable to larger areas of application, like customised mental health assessment and adaptive well-being interventions. Future research directions should be toward increasing the variability of cognitive and affective stimuli and incorporating wearable technologies in real time to enable continuous and adaptive health profiling.\u003c/p\u003e\n\u003cp\u003eIn conclusion, this study illustrates that PRAK-KNN effectively integrates machine learning-driven multimodal health profiling with traditional holistic approaches. This integration paves the way for enhanced precision in personalised and adaptive healthcare solutions.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics Approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Institutional Ethics Committee of the Indian Institute of Technology Mandi (IIT Mandi), India (Approval No. IITM/IEC(H)/2025/VD/P2). All procedures involving human participants were conducted in accordance with the ethical standards of the institutional research committee and the 1964 Helsinki declaration and its later amendments. Informed written consent was obtained from all participants prior to their inclusion in the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are not publicly available due to participant confidentiality constraints but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe code used for preprocessing, feature extraction, and model training is available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eV.D. conceived the study, supervised the research, and contributed to manuscript writing. A.B. assisted in experimental design and guided multimodal data integration. N.B. implemented the machine learning pipeline, performed the data analysis, and led manuscript drafting. A.S., G.C., R.S., and A.C. contributed to data collection and organisation. All authors reviewed the manuscript and approved the final version for submission.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to acknowledge the financial support of the All India Council for Technical Education\u0026apos;s grant to Prof. Varun Dutt (PI). Also, we would like to thank the Indian Institute of Technology Mandi for helping us with computational resources for the project. \u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDash, S., Sharma, M., Mitra, P. Multimodal biometric systems: An overview. Pattern Recognition 93, 379\u0026ndash;393 (2020).\u003c/li\u003e\n\u003cli\u003eRoy, S., Bhattacharyya, D., Kim, T.H. Multimodal biometric fusion techniques: A review. Int. J. Comput. Inf. Syst. Ind. Manag. Appl. 12, 235\u0026ndash;245 (2020).\u003c/li\u003e\n\u003cli\u003eD\u0026rsquo;Angelo, S., Eickhoff, S.B. Virtual reality and neurocognitive assessment: Toward smarter models. Front. Psychol. 11, 605925 (2020).\u003c/li\u003e\n\u003cli\u003eMadole, A., Joshi, A., Bharath, R.D. Virtual reality paradigms for assessment of cognitive functioning: A review. Sci. Rep. 12, 4650 (2022).\u003c/li\u003e\n\u003cli\u003eZhao, X., Zhang, L., Yang, Z. Combining multimodal physiological signals for emotion recognition using ensemble methods. IEEE Trans. Affect. Comput. 12, 1108\u0026ndash;1121 (2021).\u003c/li\u003e\n\u003cli\u003eJuyal, R.C., Mishra, M., Nautiyal, R., Prasher, B. Ayurgenomics: A new approach in personalized and preventive medicine. Sci. Cult. 78, 478\u0026ndash;482 (2012).\u003c/li\u003e\n\u003cli\u003eJoshi, R., Bhatt, N., Ghodke, Y., Patwardhan, B. Ayurvedic concept of Prakriti and its scientific validation through genetic markers. Indian J. Med. Res. 149, 35\u0026ndash;41 (2019).\u003c/li\u003e\n\u003cli\u003eSharma, H., Chandola, H.M., Singh, G., Basisht, G. Genetic basis of Ayurveda: Ayurgenomics. Evid. Based Complement. Alternat. Med. 2015, 1\u0026ndash;7 (2015).\u003c/li\u003e\n\u003cli\u003eShankar, D., Tiwari, R., Kumar, P. Validation of Prakriti-based Ayurvedic diagnostics using modern scientific tools. Front. Physiol. 11, 585960 (2020).\u003c/li\u003e\n\u003cli\u003eMishra, M., Prasad, R., Tiwari, A. Facial analysis for Prakriti classification using machine learning. J. Ayurveda Integr. Med. 8, 60\u0026ndash;65 (2017).\u003c/li\u003e\n\u003cli\u003eBidve, P., Mishra, S., Annapurna, J. Enhancing Ayurvedic diagnosis using multinomial naive Bayes and K‑modes clustering. arXiv:2310.02920 (2023).\u003c/li\u003e\n\u003cli\u003eHeartMath Institute. Heart rate variability and biofeedback research. HeartMath Research Library (Accessed 2024).\u003c/li\u003e\n\u003cli\u003eMuse. Muse S: The sleep and meditation headband. Interaxon Inc. (Accessed 2024).\u003c/li\u003e\n\u003cli\u003eMehta, P., Singh, A., Mehra, S. Virtual reality in Ayurvedic cognitive assessment: A pilot study. J. Tradit. Complement. Med. 12, 240\u0026ndash;251 (2022).\u003c/li\u003e\n\u003cli\u003eRotti, H., Raval, R., Anchan, S., et al. Determinants of Prakriti and its correlation with genomic variability. BMC Complement. Altern. Med. 14, 1\u0026ndash;10 (2014).\u003c/li\u003e\n\u003cli\u003eBhushan, P., Kalpana, J., Arvind, C. Classification of human population based on HLA gene polymorphism. J. Altern. Complement. Med. 11, 349\u0026ndash;353 (2005).\u003c/li\u003e\n\u003cli\u003eInteraxon Inc. Muse S‑band EEG system User Manual (2024).\u003c/li\u003e\n\u003cli\u003eSmith, J.A., et al. Cognitive correlates of EEG frequency bands: A review. Neurosci. Biobehav. Rev. 50, 194\u0026ndash;203 (2015).\u003c/li\u003e\n\u003cli\u003eHeartMath LLC. EmWave Pro Personal Stress Reliever Guide (2023).\u003c/li\u003e\n\u003cli\u003ePavlidis, I., et al. Thermal imaging for stress detection: A review. IEEE Trans. Affective Comput. 3, 102\u0026ndash;115 (2012).\u003c/li\u003e\n\u003cli\u003eKazemi, V., Sullivan, J. One millisecond face alignment with ensemble of regression trees. Proc. CVPR, 1867\u0026ndash;1874 (2014).\u003c/li\u003e\n\u003cli\u003ePatil, R., et al. Tongue image analysis in health screening: A survey. Comput. Med. Imaging Graph. 58, 35\u0026ndash;45 (2017).\u003c/li\u003e\n\u003cli\u003eHan, J., et al. Gait recognition via CNNs: A comprehensive study. Pattern Recognit. 87, 1\u0026ndash;15 (2019).\u003c/li\u003e\n\u003cli\u003eCoSpaces Edu. Virtual Reality Storytelling Platform Documentation (2024).\u003c/li\u003e\n\u003cli\u003eChatterjee, S., et al. King Harishchandra narrative in cognitive VR tests. J. Virtual Real. Med. 5, 45\u0026ndash;53 (2021).\u003c/li\u003e\n\u003cli\u003eSanei, S., Chambers, J.A. EEG Signal Processing. Wiley (2007).\u003c/li\u003e\n\u003cli\u003eNiedermeyer, E., da Silva, F.L. Electroencephalography: Basic Principles, Clinical Applications. Lippincott (2005).\u003c/li\u003e\n\u003cli\u003eShaffer, F., Ginsberg, J.P. Heart Rate Variability Metrics and Norms. Front Public Health, 5, 258 (2017).\u003c/li\u003e\n\u003cli\u003eIoannou, S., Gallese, V., Merla, A. Thermal imaging in psychophysiology. Psychophysiology, 51(10), 951\u0026ndash;963 (2014).\u003c/li\u003e\n\u003cli\u003eCao, Z., Simon, T., Wei, S.E., Sheikh, Y. Realtime Pose Estimation using PAF. Proc. CVPR, 1302\u0026ndash;1310 (2017).\u003c/li\u003e\n\u003cli\u003eSagonas, C., Tzimiropoulos, G., Zafeiriou, S., Pantic, M. Facial landmark annotation. Proc. CVPR Workshops, 896\u0026ndash;903 (2013).\u003c/li\u003e\n\u003cli\u003eField, A. Discovering Statistics Using IBM SPSS Statistics (4th ed.). Sage (2013).\u003c/li\u003e\n\u003cli\u003eBreiman, L. Random Forests. Machine Learning, 45(1), 5\u0026ndash;32 (2001).\u003c/li\u003e\n\u003cli\u003eFriedman, J.H. Greedy Function Approximation. Annals of Statistics, 29(5), 1189\u0026ndash;1232 (2001).\u003c/li\u003e\n\u003cli\u003eGeurts, P., Ernst, D., Wehenkel, L. Extremely randomized trees. Machine Learning, 63(1), 3\u0026ndash;42 (2006).\u003c/li\u003e\n\u003cli\u003eChen, T., Guestrin, C. XGBoost. Proc. ACM SIGKDD, 785\u0026ndash;794 (2016).\u003c/li\u003e\n\u003cli\u003eHosmer, D.W., Lemeshow, S. Applied Logistic Regression (2nd ed.). Wiley (2000).\u003c/li\u003e\n\u003cli\u003eCortes, C., Vapnik, V. Support-vector networks. Machine Learning, 20(3), 273\u0026ndash;297 (1995).\u003c/li\u003e\n\u003cli\u003eCover, T., Hart, P. Nearest neighbor pattern classification. IEEE Trans. Inf. Theory, 13(1), 21\u0026ndash;27 (1967).\u003c/li\u003e\n\u003cli\u003eWolpert, D.H. Stacked Generalization. Neural Networks, 5(2), 241\u0026ndash;259 (1992).\u003c/li\u003e\n\u003cli\u003eKohavi, R. Cross-validation and bootstrap. Proc. IJCAI-95, 1137\u0026ndash;1143 (1995).\u003c/li\u003e\n\u003cli\u003eBergstra, J., Bengio, Y. Random Search for Hyper-Parameter Optimization. JMLR, 13, 281\u0026ndash;305 (2012).\u003c/li\u003e\n\u003cli\u003eBishop, C.M. Pattern Recognition and Machine Learning. Springer (2006).\u003c/li\u003e\n\u003cli\u003eKshirsagar A., Rao P., Patwardhan B. Systems biology and Ayurveda. J. Ayurveda Integr. Med. 12(2), 187\u0026ndash;194 (2021).\u003c/li\u003e\n\u003cli\u003eMishra M., Prasad R., Tiwari A. Facial analysis for constitution classification. J. Ayurveda Integr. Med. 8(1), 60\u0026ndash;65 (2017).\u003c/li\u003e\n\u003cli\u003eUdupa K., Sathyaprabha T.N., Thirthalli J. HRV in constitution classification. J. Ethnopharmacol. 234, 48\u0026ndash;52 (2019).\u003c/li\u003e\n\u003cli\u003eD\u0026rsquo;Angelo S., Eickhoff S.B. VR and neurocognitive assessment. Front. Psychol. 11, 605925 (2020).\u003c/li\u003e\n\u003cli\u003eWilson M. Six views of embodied cognition. Psychon Bull Rev. 9(4), 625\u0026ndash;636 (2002).\u003c/li\u003e\n\u003cli\u003ePatwardhan B., Bodeker G., Vaidya A.D.B. Ayurvedic genomics. J. Altern. Complement. Med. 21(6), 340\u0026ndash;344 (2015).\u003c/li\u003e\n\u003cli\u003eDietterich T.G. Ensemble methods in machine learning. In MCS Proc., 1\u0026ndash;15 (2000).\u003c/li\u003e\n\u003cli\u003eLuong, T., L\u0026eacute;cuyer, A., Martin, N., Argelaguet, F. Affective and Cognitive VR. IEEE TVCG 28, 5154\u0026ndash;5171 (2021).\u003c/li\u003e\n\u003cli\u003eLi, M., et al. Emotion-Cognition Interaction in VR. IEEE TVCG 30, 2044\u0026ndash;2054 (2024).\u003c/li\u003e\n\u003cli\u003eBhushan, P., Kalpana, J., Arvind, C. Prakriti classification by HLA polymorphism. J. Altern. Complement. Med. 11, 349\u0026ndash;353 (2005).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"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":"Personalised health profile, machine learning, multimodal biometrics, physiological indications, cognitive evaluation, virtual reality environments","lastPublishedDoi":"10.21203/rs.3.rs-7044176/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7044176/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCorrect classification of individualised health profiles is paramount for the progress of precision health strategies but is difficult for conventional constitution-based systems based on subjective judgments. This work fills this void by suggesting a machine learning paradigm that objectively characterises individual health types based on multimodal physiological and neurocognitive markers. Heart rate variability (HRV), electroencephalography (EEG), gait dynamics, facial thermography, facial imaging, and cognitive-emotional responses in a virtual reality (VR) environment were obtained from 48 subjects. Feature selection, an integration of analysis of variance (ANOVA) and Random Forest-based importance ranking, minimised the dataset from 214 to 69 key variables. A novel meta ensemble model, PRAK-KNN, using multiple machine learning (ML) algorithms, XGBoost and Extra Trees for adaptive neighbor weighting, achieved an accuracy in classification of up to 76% with different feature subsets. In virtual reality-based tests, cognitive load and emotional arousal were key discriminators among individual profiles. This study illustrates an objective, replicable method for constitution-type classification, opening the door to incorporating classical health ideas into contemporary data-driven personalised medicine paradigms.\u003c/p\u003e","manuscriptTitle":"Machine Learning Framework for Individualised Health Profiling Using Multimodal Physiological and Neurocognitive Biomarkers","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-23 07:29:19","doi":"10.21203/rs.3.rs-7044176/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":"cd42553d-9315-4472-ba8c-dbd227e54938","owner":[],"postedDate":"July 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":51711793,"name":"Health sciences/Biomarkers"},{"id":51711794,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":51711795,"name":"Physical sciences/Engineering"},{"id":51711796,"name":"Health sciences/Health care"},{"id":51711797,"name":"Physical sciences/Mathematics and computing"},{"id":51711798,"name":"Biological sciences/Neuroscience"}],"tags":[],"updatedAt":"2025-08-04T10:39:08+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-23 07:29:19","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7044176","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7044176","identity":"rs-7044176","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}