Machine learning-based risk classification of depressive symptoms among patients with hearing loss: evidence from the Health and Retirement Study (HRS).

Pain is a multifaceted and subjective phenomenon frequently experienced by patients in intensive care units. In non-communicating populations, conventional assessment tools are often inadequate and susceptible to observer bias. Deep learning-based facial analysis has emerged as a promising approach for the objective quantification of observable pain-related behavioural indicators. To evaluate the feasibility and diagnostic accuracy of deep-learning models in categorising pain severity in non-communicative adult intensive care patients, using expert-annotated facial images. Features were extracted via the DenseNet-169 architecture, dimensionally reduced with principal component analysis and classified using support vector machine, random forest and K-nearest neighbours. Data sets were independently annotated by a multidisciplinary team comprising an intensivist, intensive care nurses and a pain specialist. Model performance was comprehensively assessed through accuracy, precision, sensitivity, the F1 score, the area under the receiver operating characteristic curve and Fleiss' kappa coefficient to ensure robust inter-rater reliability. A total of 636 facial images obtained from 120 adult intensive care unit patients were analysed. The support vector machine model achieved the highest overall performance, with an accuracy of 96.9% and an area under the receiver operating characteristic curve of 0.994, demonstrating exceptional sensitivity in severe pain classification. While K-nearest neighbours showed superior performance for moderate pain detection, random forest yielded the lowest accuracy across all data sets. Notably, inter-rater agreement was low (k = 0.16), highlighting the significant variability in expert human judgements and the subjective nature of manual pain assessment. Deep learning-based facial analysis provides a valid, reproducible and standardised method for pain assessment in non-verbal intensive care patients. The creation of a multi-expert annotated data set and the systematic comparison of classifiers across diverse clinical perspectives represent the original contributions of this study. Automated facial expression analysis minimises inter-observer variability by providing an objective decision support mechanism for critical care nurses. This technology facilitates the standardisation of pain management protocols and bolsters patient safety by reducing the inherent risks of subjective assessment bias.

This study aimed to identify and characterize behavioral subtypes of potentially problematic gambling patterns using real-world data from a major Danish gambling provider. A two-stage clustering approach was applied to behavioral data from over 650,000 users. In the first stage, k-means clustering segmented the population based on standardized behavioral indicators. In the second stage, subtypes were identified within selected clusters using k-medoids clustering. The procedure was replicated in an independent dataset (N= 619,441) to assess structural stability. The analysis revealed clear behavioral heterogeneity across gambling profiles. Subtypes differed systematically in intensity, reloads, product focus, and risk-related patterns such as chasing and nighttime gambling. The overall cluster structure was consistent across datasets, supporting the robustness and generalizability of the identified subtypes. Gambling-related risk patterns does not follow a single behavioral pattern. The data-driven subtypes of potentially problematic gambling identified in this study highlight the heterogeneity of gambling-related risk and offer a foundation for more nuanced approaches to risk detection based on gambling patterns.

Predicting student academic performance has become increasingly vital in the field of educational data mining, as institutions seek data-driven strategies to enhance learning outcomes. However, many existing models rely solely on behavioral indicators or static features, often overlooking the role of time and context in shaping learning behavior. This limitation reduces predictive accuracy and adaptability in academic environments. To address this challenge, this study introduces EduFuseNet, a hybrid deep learning framework that integrates behavioral and spatiotemporal data for accurate classification of student performance. The workflow begins with data collection from a Student Academic Performance dataset, comprising both behavioral metrics and spatiotemporal information. The raw data undergoes preprocessing, including missing value imputation, one-hot encoding of categorical variables, and min-max scaling of numerical features. The processed data is then passed through two specialized branches: a Tabular Neural Structure-Aware (TabNSA) module that captures complex interdependencies within behavioral data, and a Spatiotemporal Transformer module that models temporal and sequential patterns in learning activities. The feature embeddings from both branches are fused and passed through fully connected layers to generate predictions across five academic performance bands, enabling precise classification and early risk identification. EduFuseNet achieved an accuracy of 99.00%, with a precision of 99.04%, recall of 99.00%, and F1-score of 99.01%, reflecting strong and reliable predictive performance. By leveraging both behavioral and temporal learning indicators, the model serves as an effective tool for early academic monitoring and intervention.

Emotions play an important role in shaping aggressive behavior, and understanding their underlying psychological mechanisms is particularly relevant among college students. However, existing research has predominantly focused on reactive aggression, while comparatively less attention has been paid to proactive aggression, which is more instrumental in nature and associated with more severe social consequences. In addition, empirical evidence regarding the valence-specific effects of awe remains limited. The present study aimed to examine the differential effects of positive and negative awe on proactive aggression and to explore the role of empathy as a potential mediating mechanism. A total of 110 college students were randomly assigned to one of three conditions: positive awe, negative awe, or neutral emotion. Awe was induced through video clips depicting natural landscapes. Proactive aggression was assessed using a modified bug-killing paradigm, including two behavioral indicators: force intensity and proportion of bugs killed. Empathy was measured using the Interpersonal Reactivity Index. The results revealed a clear differentiation based on the valence of awe. Participants in the positive awe condition exhibited significantly lower levels of proactive aggression than those in the neutral condition across both force intensity (M = 2.86, SD = 0.81 vs. M = 4.17, SD = 0.81) and proportion of bugs killed (M = 0.68, SD = 0.25 vs. M = 0.93, SD = 0.11). In contrast, the inhibitory effects of negative awe were weaker and less consistent. Compared with the neutral condition, negative awe was associated with a lower proportion of bugs killed, although this effect only reached marginal significance (p = 0.06, η2 = 0.04), and no significant difference was observed for force intensity. Mediation analyses indicated that empathy partially mediated the association between positive awe and proactive aggression. Empathy accounted for 31% of the total effect in the force intensity pathway (B = −0.02, t = −4.25, p < 0.001, 95% CI [−0.04, −0.01]) and 18% in the proportion-of-bugs-killed pathway (B = −0.003, t = −2.37, p = 0.02, 95% CI [−0.006, −0.001]). Notably, no significant mediating effect of empathy was observed in the negative awe condition, suggesting that the psychological processes linking awe to proactive aggression may differ as a function of emotional valence. Taken together, the present findings suggest that positive awe is reliably associated with lower levels of proactive aggression among college students, and that this association is partially explained by increased empathy. By contrast, the effects of negative awe appear to be fragile and context-dependent, as reflected in their failure to reach statistical significance, indicator-specific manifestation, and the absence of a consistent mediating pathway. These results highlight the importance of distinguishing between positive and negative awe when examining the behavioral consequences of self-transcendent emotions and underscore the need for further research to clarify the conditions under which negative awe may influence aggressive behavior.

Objectives: This study aims to develop an optimization based framework for predicting student academic performance using smartphone behavioural indicators. Method: The proposed Margin-based Rule Learning Integrated Network with eXplainable Optimization (MARLIN-X) framework integrates Tri-Label Margin Feature Selection, DCAR-Net classification and Rule-Set Distillation under a novel Mosaic Patch Migration Optimizer. The optimizer operates using patch based population structuring, migration, fragmentation and elite gene preservation. Feature subsets, hyperparameters and rule masks are optimized within a unified search process. Findings: The proposed method achieves 97.8% accuracy on the School dataset and 98.3% on the College dataset. Feature dimensionality is reduced by more than 50% while increasing tri-label margin values. The distilled rule set achieves 96.5% fidelity with respect to deep model predictions. Balanced accuracy and Macro-F1 scores show consistent improvement over recent methods. Novelty: The study introduces a Mosaic Patch Migration Optimizer that performs patch-level structured search rather than individual-level movement. Keywords: Academic Performance Prediction, Smartphone Usage Analytics, Feature Selection, Nature-Inspired Optimization, Deep Learning, Rule Distillation, Educational Data Mining

ABSTRACT Background to the Study As fully online postgraduate programmes expand, questions remain regarding whether sufficient student engagement is achieved and how such sufficiency can be measured. This study examined the types and levels of engagement within a fully online postgraduate module and explored how engagement can be operationalised using learning management system (LMS) analytics. Objective To explore whether there is sufficient student engagement in an online module, and the types and levels of online engagement. Methods A quantitative single‐case study analysed LMS trace data from 773 students. Data were analysed using the Online Engagement Framework and Moore's interaction typology. Engagement was operationalised using four behavioural indicators: submissions, interactions, time‐on‐platform and Grade Center access. Cluster analysis was applied to identify engagement profiles. Results Findings indicate high levels of social, cognitive, behavioural and collaborative engagement, with participation substantially exceeding minimum requirements. In contrast, structured opportunities for emotional engagement were absent. Frequent Grade Centre access (mean = 68 views per student) suggests a digitally observable form of performance engagement characterised by academic self‐monitoring behaviour Cluster analysis revealed four distinct engagement profiles, highlighting heterogeneity in student interaction patterns. Conclusion The findings suggest that high‐density programmatic assessment is associated with sustained engagement behaviours in online contexts. This study contributes to the literature by proposing a trace‐based operationalisation of performance engagement and offering a practical framework for examining engagement sufficiency in fully online programmes.

Autistic populations are more likely to need healthcare (HC) services due to co-occurring mental health issues, including anxiety and attention-deficit hyperactivity disorders. One of the most significant barriers to delivering optimal medical procedures in autism spectrum disorder (ASD) is the sensory overload in HC settings. Divergent sensory processing, along with unpredictable built environments (BEs), can exacerbate stress-induced anxiety and avoidant behaviour. A growing body of systematic studies links autism-friendly BEs with positive care experiences, yet substantial gaps remain in understanding the effects on behavioural and physiological aspects of emotional responses. This systematic review aims to comprehensively evaluate the HC-BE features that impact on behavioural indicators and non-invasive biomarkers of stress, anxiety and sensory processing in patients with ASD, to establish best practices. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines will be followed. Peer-reviewed articles in PubMed, Embase, Web of Science, Scopus, and PsycINFO databases and Google Scholar will be searched. Studies will be selected if they apply qualitative, quantitative, or mixed-methods designs. Two independent reviewers will select studies at the title and abstract, and full-text screening stages. Data will be extracted by one reviewer and verified by review members using a crowdsourcing approach for quality assurance. Risk of bias will be assessed by one reviewer using the Cochrane Risk of Bias Tools, The Critical Appraisal Skills Programme, and The Mixed Methods Appraisal Tool, and checked by the reviewer with methodological expertise. A results-based convergent synthesis design is planned for data synthesis. This review, which converges indicators and patient experiences, will provide a complete overarching picture of the inherent complexities associated with HC-BE and autistic individuals. The findings can inform decisions and recommendations for research and practice. PROSPERO CRD42024562288.

Mental health problems tied to negative affective experiences are common among emerging adults, yet conventional symptom questionnaires provide only distal snapshots of dynamic affective processes that unfold in daily life. Using the novel Meet Pandora smartphone application, we examined how everyday affective experiences and passive markers relate to depressive symptoms (PHQ), anxiety symptoms (GAD), and psychological flourishing in college students ( N = 120; N observations (PHQ/GAD) = 372, N observations (flourishing) = 792). We tested three data channels (self-report, voice-derived features, and behavioral indicators) and disaggregated within-person fluctuations from between-person differences using multilevel models. Across outcomes, the proportion of positive emojis selected to reflect one's affective state emerged as the most consistent signal. On days when participants used a higher-than-usual proportion of positive emojis, they reported lower depression and anxiety, and individuals with higher average positive emoji use reported lower symptoms and higher flourishing. Positive emoji use also moderated change over time in flourishing, such that flourishing increased across the study period only among participants with relatively high positive emoji proportions. In contrast, passive features showed more selective associations. Speech rate (words per minute) was linked to lower symptom burden and higher flourishing in some models, and longer average sleep duration was associated with lower anxiety and higher flourishing. Overall, results highlight the value of separating within- from between-person effects when linking digital markers to mental health and suggest that low-burden indicators of positive affect may be especially informative for scalable temporal monitoring of affective risk and wellbeing in young adults.

Context Residual feed intake (RFI) and residual intake and gain (RIG) are widely used metrics to assess feed efficiency in sheep. However, their practical implementation in breeding programs remains limited owing to the high cost and complexity of individual feed intake measurements. Consequently, alternative indicators, such as physiological responses and infrared thermography (IRT), have been proposed as potential proxies for identifying animals with superior feed efficiency under different environmental conditions. Aims This study aimed to evaluate the relationships between RFI and RIG classifications and physiological parameters, as well as surface body temperatures obtained via IRT, in Texel ewes exposed to natural heat stress. Methods Thirty-nine young Texel ewes were monitored for 57 days in a covered facility equipped with an automated feeding and watering system with individual intake recording. Animals were classified into low, medium, and high-efficiency groups according to their RFI and RIG values. Physiological responses, including respiratory rate (RR), heart rate (HR), rectal temperature (RT), and the heat tolerance coefficient (HTC), were recorded. Additionally, surface temperatures of the eye, muzzle, hooves, and vulva were measured using infrared thermography (IRT). Statistical analyses included ANOVA and principal component analysis (PCA) to explore associations among traits. Key results No significant differences were detected among RFI or RIG classes for RR, HR, RT, or HTC. Similarly, IRT-derived surface temperatures did not differ across efficiency classifications. PCA showed that RR and HTC explained the greatest proportion of total variance, whereas RFI and RIG contributed to other independent components. Conclusions Neither physiological parameters nor IRT-based surface temperatures were effective indicators of feed efficiency in Texel ewes under natural heat stress. Implications The results indicated that RFI and RIG cannot be accurately inferred from physiological or IRT variables under field conditions. Future research should integrate additional phenotypic and behavioral indicators to identify reliable, low-cost biomarkers for metabolic efficiency in sheep.

This study addresses the critical need for soft skills development among orphaned and underprivileged youth, focusing on the persistent challenge of public speaking anxiety and low self-confidence. The objective was to design, implement, and evaluate a targeted training program to enhance self-efficacy and foundational public speaking abilities for students at the Griya Yatim & Dhuafa (GYD) BSD 3 orphanage. The methodology involved a one-day experiential workshop titled "Speak Up! I Can!", integrating interactive theoretical instruction on communication principles with practical simulation exercises. A direct observational approach was used to assess changes in key behavioral indicators of confidence before and after the intervention. Results demonstrated significant positive shifts: participant willingness to speak publicly increased dramatically to approximately 80%, accompanied by marked improvements in non-verbal communication cues such as eye contact, posture, and facial expression. The overall estimated confidence score for the group rose from a low to a moderate range. The program concluded that a structured, practice-oriented workshop grounded in self-efficacy theory is an effective intervention for initiating confidence and skill development in vulnerable youth populations. This initiative contributes to the field by demonstrating the practical application of communication science in community empowerment and offers a replicable model for similar interventions aimed at enhancing the psychosocial readiness of at-risk children.

Immersion anaesthesia is widely used in fish handling and research, yet optimization of protocols requires not only effective immobilization but also consideration of internal physiological responses. This study evaluated the effects of immersion-administered lidocaine (60 and 80 mg/L) as an adjuvant to clove essential oil (CEO; 100 μL/L) on anaesthesia induction and recovery in juvenile Oreochromis niloticus under tightly controlled temperature (27.5°C ± 0.3°C). Behavioural endpoints included loss of equilibrium (LOE), time to an operational anaesthetic plane and recovery time; opercular respiratory rate was assessed as an internal physiological indicator. In the main experiment, adding lidocaine to CEO did not significantly alter LOE, time to anaesthetic plane or recovery time relative to CEO alone. In contrast, respiratory-rate trajectories differed significantly among treatments during both induction and recovery, indicating protocol-dependent modulation of ventilatory dynamics despite similar time-based anaesthetic endpoints. A complementary preliminary experiment showed that lidocaine alone (60 or 80 mg/L) did not induce LOE or anaesthetic plane within a predefined 5-min exposure window, although it produced clear dose-dependent changes in respiratory rate relative to water and ethanol controls. Together with qualitative observations of reduced behavioural activation, these findings are compatible with a possible calming or sedation-like effect of lidocaine, although this was not formally measured and should be considered a hypothesis. Overall, lidocaine did not improve the operational speed of a CEO-based immersion anaesthetic protocol in Nile tilapia, but it did modulate physiological responses relevant to protocol refinement. These results highlight the importance of integrating both behavioural and internal physiological indicators, together with strict thermal control, when evaluating immersion anaesthesia protocols in fish.

Student focus is a significant element that defines the effectiveness of the learning process, but conventional methods of observation are also subjective and ineffective in cases of virtual learning or when there are many students in the classroom. The proposed project is a real-time student attention monitoring system, the implementation of which is based on deep learning and computer vision, which will automatically identify and analyze student engagement. The system employs a regular web camera to monitor behavioral indicators like openness of the eyes, mouth motions (or speaking or yawning) and the orientation of the head to categorizeeach student into being engaged or not. It integrates face detection of Haar Cascades and head pose estimation of deep learning to provide consistency during work in varying lightning and facial features. Also, in cases whereby a student is realized not to be attentive, the system captures and records the image of the student in real time, and the image is stored safely in the teacher login where he can access and track the student later. Attentive cases are also monitored by the system, and it also offers insights in the form of an interactive dashboard that shows summaries of the engagement and real-time graphs to aid in assessing the effectiveness of the teaching. It is a scalable solution that is designed to be low-priced, non-invasive, and only slightly involves the teacher, minimizing human bias, and encouraging improved engagement and education (physical classroom or online). On balance, the project shows that it is possible to teach smart and data-driven through the application ofdeep learning, which results in better teaching performance and student engagement.

In the context of the educational reform under the 2018 General Education Curriculum, designing competency-oriented exercises for the Natural and Social Sciences subject is an urgent requirement but remains a major challenge for many primary school teachers. This article presents a scientific 4-step exercise design process aimed at developing "the competency of exploring the surrounding natural and social environment" for students through a case study on the theme "Family" in Grade 2. Based on the analysis of theoretical foundations and practical surveys, we have specified this competency into 3 core behavioral indicators (T1, T2, T3) and successfully designed a system of 54 exercises associated with practical situations, accompanied by a 3-level evaluation Rubric. To verify its effectiveness, a quasi-experimental study was conducted on 57 Grade 2 students (including 28 students in the experimental group and 29 students in the control group) in Da Nang city. The results of the qualitative process evaluation via Radar charts show the progressive and steady improvement of the experimental group. Quantitatively, the average score of the comprehensive test of the experimental group reached 7.64 points, significantly higher than the 7.03 points of the control group; in particular, the rate of Good - Excellent scores in the experimental group reached 50% compared to 31.03% in the control group. The research results affirm that the proposed process and exercise system have high feasibility, resolve the difficulties regarding learning materials for teachers, and contribute to substantially improving the teaching quality of the Natural and Social Sciences subject at the primary school level.

In increasingly competitive business environments, maintaining customer loyalty has become a critical factor for sustaining long-term organizational performance. However, in many small and medium-sized enterprises, identifying loyal customers is still conducted subjectively, leading to inconsistent, non-transparent, and potentially biased reward allocation decisions. This study proposes a data-driven framework for customer loyalty ranking by integrating a Simple Additive Weighting (SAW)-based decision support approach. The research adopts a quantitative applied methodology using transactional data from a furniture retail business, covering a period of 12 months and involving 1,000 customer transactions. Customer loyalty is evaluated based on three key criteria: monetary value, purchase frequency, and payment reliability, which represent essential behavioral indicators of customer engagement. The SAW method is employed to normalize criteria values, assign relative weights, and compute preference scores for each customer, resulting in a systematic and objective ranking process. The proposed framework is implemented as a web-based decision support system using PHP with the CodeIgniter framework and a MySQL database to ensure structured data management and operational efficiency. The results demonstrate that the framework effectively produces consistent, transparent, and data-driven customer rankings, thereby reducing subjectivity in managerial decision-making. This study contributes by formalizing a practical decision support framework that enhances the reliability, fairness, and effectiveness of customer loyalty evaluation and reward allocation, offering a novel integration of data-driven decision-making paradigms with the SAW-based decision support system, a feature often underexplored in prior studies.

The large yellow croaker (Larimichthys crocea) is a high-value marine fish, but stress during live transport often leads to physiological disturbance and deterioration of muscle quality. This study investigated the effects of pre-transport temporary rearing at three temperatures (8, 10, and 12 °C) over 48 h on stress response, energy allocation, and muscle quality in this fish species. Temporary rearing at 8 °C induced stronger cold stress, characterised by elevated cortisol, marked lipid mobilisation, late lactate rebound, and greater loss of polyunsaturated fatty acids, indicating enhanced stress–catabolism coupling and higher risk of quality deterioration. In contrast, 12 °C did not sufficiently suppress metabolic turnover, resulting in continuous glycogen depletion, rapid ATP degradation, and accelerated accumulation of bitter-tasting nucleotide metabolites such as hypoxanthine. Among the tested temperatures, 10 °C showed the most coordinated response, with relatively stable endocrine status, moderate substrate utilisation, lower accumulation of undesirable degradation products, and better preservation of texture, water-holding capacity, and flavour-related precursors. These findings suggest that 10 °C is a promising pre-transport temporary rearing temperature for large yellow croakers under the present 48 h experimental conditions. The advantage of this temperature appears to lie in achieving a more favourable balance between metabolic suppression and physiological homeostasis, thereby providing a scientific basis for improving pre-transport rearing management and supporting safer, more stable live transport. Future studies incorporating behavioural and molecular indicators are needed to further clarify the regulatory effects of 10 °C during pre-transport rearing.

Fingerprints are a widely recognized form of forensic evidence, valued for their ability to link individuals with specific locations. Traditional fingerprint analysis relies on optical imaging to identify a match in a fingerprint database; however, where no match is found, the evidential value of a latent print is limited. Here, we present the first study to integrate matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS) with supervised machine learning to infer physical activity from fingerprint chemistry, expanding the utility of fingerprints beyond identification alone. Physical activity labels were derived from a validated questionnaire and converted into binary classes. Supervised machine learning algorithms were trained on the lipid features and evaluated against the survey-derived labels. The top-performing models were an ensemble algorithm based on multiple decision trees and a neural network, which classified physical activity with accuracies of 75 ± 8% and 73 ± 7%, respectively. These results demonstrate that fingerprint lipid chemistry encodes biologically meaningful information related to physical activity and establish a new approach for extracting lifestyle and behavioral indicators from trace evidence, with potential applications in forensic investigations and noninvasive fingerprint-based assessments in medicine.

ABSTRACT Early identification of academic deficits facilitates timely and effective interventions that are essential for individual development and societal progress. Conventional assessment methods provide insights through demographic and prior academic records and subjective measures; however, there is an increasing demand for objective, ecologically valid indicators that reflect real-world learning processes. Extending these approaches to younger learners, particularly those aged 11–12 years, addresses critical research gaps. This study presents an exploratory investigation of a Virtual Reality (VR)-based academic assessment tool designed to capture behavioral and performance metrics from learners. A cohort of 120 typically developing students participated, with teacher evaluations serving as the ground truth. In addition to raw VR metrics, confirmatory factor analysis (CFA) was employed to derive latent cognitive dimensions that provided a structured representation of students’ behavioral performance. Multiple machine-learning algorithms were tested to classify performance levels, with the Random Forest classifier achieving the highest accuracy of 95%. The findings demonstrate that VR-captured behavioral indicators have strong feasibility and predictive validity for educational assessment. The study also identifies a minimal, interpretable set of tasks and features that enhances practical deployment. These results provide foundational evidence for integrating VR-based assessment, latent cognitive factors, and ML methods to generate dynamic, ecologically valid insights and to support targeted early interventions for improved educational outcomes.

Rumination time is considered a sensitive behavioral indicator of physiological and metabolic status in dairy cows, yet its relationships with biochemical and milk quality parameters under commercial robotic milking conditions remain insufficiently described. This study combined precision monitoring technologies, serum biochemical profiling, and in-line milk analysis to evaluate physiological differences among early-lactation Holstein cows according to rumination time. A total of 88 cows were classified into three rumination time categories (>527, 412-527, and <412 min/day). Milk production traits, milk quality indicators, and blood biochemical parameters were compared among groups, and univariable regression analysis was performed to identify variables associated with rumination time. Cows in the low rumination group showed higher milk temperature, electrical conductivity, and somatic cell count, as well as lower milk protein percentage. They also showed higher concentrations of total protein, urea, gamma-glutamyl transferase, and lactate dehydrogenase, while triglyceride concentrations were lower. Regression analysis identified electrical milk conductivity, creatinine, magnesium, potassium, and chloride as variables associated with rumination time. These findings indicate that reduced rumination time is associated with changes in milk quality and biochemical parameters in early-lactation dairy cows, suggesting that rumination monitoring may provide useful information for identifying cows experiencing physiological and metabolic challenges under commercial farming conditions.

Attention Deficit Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental condition that significantly impairs classroom performance and academic achievement. Traditional screening relies on subjective clinical assessments, which are time-consuming and prone to observer bias. This paper presents an automated ADHD behavioral screening system that leverages computer vision and machine learning to analyze observable behavioral indicators from classroom video recordings. Eight behavioral features—yaw standard deviation, head turn ratio, average gaze deviation, sustained focus ratio, body motion energy, hand fidget frequency, mean wrist velocity, and posture ratio—are extracted per frame using MediaPipe Face Mesh and Pose estimation. A Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel and StandardScaler normalization classifies behavioral patterns as ADHD-indicative or typical. Five-fold cross-validation yields a mean accuracy of 91.4%, with an Area Under the ROC Curve (AUC) of 0.963, demonstrating the efficacy of the proposed noninvasive screening pipeline.