Jul 2026· Jurnal Informatika· pp. 353-362· 0 citations
TL;DR
Improvements in recall and F1-score for minority classes demonstrate the effectiveness of the balancing process and highlight the potential of machine learning for student mental health classification, although further validation on larger and more diverse datasets is required.
Abstract
Mental health issues among students require data-driven approaches for early identification. This study aims to classify students’ mental health levels using the Depression Anxiety Stress Scale (DASS-21) and perceived social support, measured by the Multidimensional Scale of Perceived Social Support (MSPSS), via machine learning algorithms. A supervised classification approach was employed using Random Forest, Support Vector Machine, and Logistic Regression on data collected from 450 respondents. The data were processed through scoring, labeling, encoding, balancing, and stratified 80:20 splitting. Model evaluation was conducted using hold-out testing and 5-fold cross-validation to ensure robust and reliable performance estimation. The results indicate that Random Forest achieved the best performance, with an accuracy of 0.97 on the test set, outperforming Support Vector Machine (0.81) and Logistic Regression (0.83). Improvements in recall and F1-score for minority classes demonstrate the effectiveness of the balancing process. These findings highlight the potential of machine learning for student mental health classification, although further validation on larger and more diverse datasets is required.
This technical report explains how a machine learning-based system was developed and tested to identify the risks of depression in students of a university. This research report is an answer to the demand of the proactive and easy-to-use screening tools in cases of an increasing global mental health crisis in the academic environment. The system was built on a large volume of data, 27,889 student records in total, and studied 18 major characteristics of students in terms of demographics, academic success, and mental health factors. It used a systematic 6 stage methodology that included data preprocessing, model training and performance validation. Four monitored machine learning algorithms were put into version and juxtaposed: Logistic Regression, Decision Tree, Naive Bayes, and Random Forest. The findings indicated that the Logistic Regression was the most accurate with an accuracy of 84.38 and it was higher than the other models. The main results of the analysis included that the prevalence of depression (58.5% and suicidal thoughts (63.3% are high in the sample population and indicate the severity of the issue. Suicidal Thoughts and Academic Pressure were the most important risk factors mentioned by the model with the feature importance scores of 23.2% and 17.1, respectively. The research report is able to prove the technical feasibility and clinical utility of an automated and accessible and privacy-aware screening tool. This system is not a replacement of a professional diagnosis, but it offers a helpful initial step toward recognizing at-risk students, assessing them in due time and offering the necessary assistance, and leading to more sophisticated institutional mental health approaches.
Mustafa Ahmed Abdulwahhab· Jurnal Pendidikan Kepelatiha...· 0 citations
Depression among university students has become a significant public health concern due to increasing academic pressure, financial instability, emotional stress, and social challenges. Early prediction of depression can help institutions implement timely interventions and improve student well-being. The study implements and compares four different classification algorithms: Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN). The analysis incorporates data augmentation techniques, handles class imbalance through SMOTE (Synthetic Minority Over-sampling Technique), and provides extensive model evaluation metrics, including standard train-test split evaluation and rigorous 5-fold cross-validation with aggregated confusion matrices. Data preprocessing involved handling missing values, normalization, label encoding, and feature selection. Model performance was evaluated using accuracy, precision, recall, F1-score, and Receiver Operating Characteristic Area Under Curve (ROC-AUC). Experimental results demonstrated that the Random Forest model achieves superior performance, with 91.5% accuracy and 0.904 cross-validation accuracy, significantly outperforming other architectures for predicting depression among students. The findings indicate that machine learning approaches can provide reliable tools for early depression risk identification in higher education institutions.
Adedeji Oluwaseun Bukonla· The Scientific World Journal· 0 citations
Mental health issues, especially depressive symptoms, among young adults represent a public health challenge. Conventional psychological assessment tools have limited sensitivity and specificity for identifying individuals at risk. This study aims to develop an explainable machine learning-based model to stratify concurrent depression risk in young adults. This study included 100,257 college students and collected mental health variables including depression, anxiety, resilience, parent-child relationship, and duration of mobile phone usage. The screening capabilities of 13 machine learning algorithms were systematically evaluated and compared. The SHapley Additive exPlanations (SHAP) framework was employed for the interpretability of the final model. The median scores for parent-child relationship, resilience, anxiety, and mobile phone usage time was 42.0, 28.0, 1.0 and 28.0, respectively. Among the 13 machine learning algorithms, the XGBoost model demonstrated superior performance. The final multivariate screening model achieved an area under the curve (AUC) of 0.887, a sensitivity of 0.787, a specificity of 0.830, and an accuracy of 0.816 in classifying young adults' concurrent depression risk. The SHAP analysis showed the importance of each variable: anxiety (2.303) > resilience (0.774) > parent-child relationship (0.708) > mobile phone usage time (0.411). The final multivariate model exhibited stable performance during cross-validation (AUC = 0.885 ± 0.032), significantly better than the single-variable model (P < 0.001) and better screening reliability (Brier score 0.153). The final multivariate XGBoost model provides a highly accurate and interpretable approach for young adults' depression risk stratification. As the model was developed using cross-sectional data collected during the COVID-19 campus lockdown, prospective validation is required before clinical deployment. Notably, anxiety level emerged as the most influential risk factor, and resilience demonstrated a significant protective effect.
OBJECTIVES
Cancer-related cognitive impairment is a decline in cognitive functioning following cancer treatment that negatively affects survivors' quality of life. Self-report tools such as the Cognitive Failure Questionnaire (CFQ) offer a rapid and cost-effective method for early detection of subjective cognitive difficulties. This study aimed to develop and compare machine learning models using demographic variables to classify subjective cognitive impairment (SCI) severity into low, moderate, and high levels.
METHODS
Data from 437 cancer survivors were analyzed. SCI severity was defined based on self-reported CFQ scores. Demographic predictors included age, gender, marital status, education, and occupation. Four supervised machine learning algorithms, Logistic Regression (LR), Support Vector Machine (SVM), Artificial Neural Network (ANN), and k-Nearest Neighbors (KNN), were applied for multiclass classification. Model performance was evaluated using accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1-score and area under the ROC curve (AUC), with stratified 5-fold cross-validation.
RESULTS
LR model demonstrated the best performance (accuracy = 0.89, AUC = 0.95), followed by SVM (accuracy = 0.86, AUC = 0.95), ANN (accuracy = 0.85, AUC = 0.93), and KNN (accuracy = 0.82, AUC = 0.91).
CONCLUSION
Logistic Regression provided the most robust and interpretable results. These findings support the utility of simple demographic-based models as noninvasive tools that may support early supportive screening of subjective cognitive difficulties in cancer survivorship care. However, external validation using independent datasets is required.
Mental health disorders such as anxiety, depression, bipolar disorder, personality disorders and stress have become significant public health concerns. This paper proposes a comparative analysis of machine learning models based on Explainable Artificial Intelligence (XAI) for mental health diagnostics using social media data from Reddit and Twitter. Two publicly available mental health related datasets were preprocessed using tokenization, stopwords removal, removal of special characters, removal of URLs, and TF-IDF vectorization. Logistic Regression, Naïve Bayes, Decision Tree, and Random Forest classifiers were analyzed before and after XAI-based LIME technique was applied. LIME technique helped to determine influential features and increase transparency of the models using feature selection. From experimental results, it can be concluded that Logistic Regression provided the best accuracy (76.84%) among all classifiers on the largest dataset after applying LIME-based feature refinement, while Decision Tree and Random Forest provided 81.94% accuracy on the second dataset using positive features.
Swarnali Daw, Upasana Saha, Suchismita Maiti et al.· 2026 4th International Confe...· 0 citations
Student mental health has a direct bearing on academic performance and overall well-being, yet problems such as
stress, anxiety and depression usually go unnoticed until they have already affected a student's grades or attendance.
Counsellors and academic staff typically rely on manual observation, self-reported questionnaires, or referrals to identify
students who may be struggling, an approach that is reactive rather than preventive and does not scale well across large student
populations. This paper presents MindTrack, a machine learning based framework for predicting student mental health risk
from a combination of academic and psychological indicators, namely age, gender, study hours, sleep hours, attendance, CGPA,
stress level, anxiety level and depression level. Five supervised classifiers — Support Vector Machine (SVM), Random Forest,
Decision Tree, K-Nearest Neighbors (KNN) and Naive Bayes — were trained on a dataset of 12,000 student records and were
independently validated on a separate set of 399 records so that the reported performance reflects behaviour on unseen data
rather than the training data itself. Among the five models, the Support Vector Machine produced the most balanced and
reliable results, reaching a validation accuracy of 80.45% and an F1-score of 0.8203. A Random Forest based feature
importance analysis further showed that stress level, anxiety level and depression level were the strongest predictors of risk,
together accounting for more than 71% of the total feature contribution. All five trained models were deployed inside a Flask
based web dashboard that walks a user through an Overview, Academic Info, Health Indicators and Results screen, returning an
instant High Risk / Low Risk classification along with a side-by-side comparison of every model and a feature importance chart.
The results indicate that combining routinely available academic records with short psychological self-assessments can support
earlier and more consistent identification of at-risk students than manual counselling alone.
Mohammed Aashiq Farhaan, Dr. T. Siva Rama Krishna· International Journal for Re...· 0 citations
