Research, cited.

Suicide among adolescents is an escalating worldwide public health problem costing more than 800,000 lives every year and affecting low and middle-income countries (LMICs) disproportionately, where mental health resources are often limited. Conventional clinical screenings are time consuming and can be error-prone. We propose an innovative machine learning framework with privacy-preserving algorithms based on standardized Global School Based Student Health Survey data from 4 LMICs (Bangladesh 2014, Nepal 2015, Thailand 2015, Timor-Leste 2015) (total 19116). Suicidal behaviors (ideation, planning and attempts) was combined into a binary risk indicator. K-Nearest Neighbors imputer was used for handling the missing data followed by Synthetic Minority Over-Sampling Technique (SMOTE) for addressing the problem of class imbalance. This is for the seven models which were evaluated, and Extra Trees had the best overall performance with 98.04% accuracy on the Bangladeshi cohort and with good generalizability across the multi-country pooled dataset (accuracy 92.89%, F1 0.928, AUC 0.977). Through feature analysis, five highly stable features were identified: lack of close friendships, bullying, loneliness, sleep disturbance, and early sexual activity. This framework offers an ethical and evidence-based approach to risk screening for adolescents below age 16 experiencing suicidal ideation and/or suicidal behaviors in resource-poor communities.Cite

Student academic prediction has been extensively explored but no systems are available that explain why a student underperforms, and what specific action steps might help improve outcomes. We propose FairCF, a hybrid explainable AI framework that integrates XGBoost-based student performance prediction, DiCE-based counterfactual explanation generation, and demographic fairness auditing to provide interpretable and fairness-aware decision support for educational analytics. We further incorporate a demographic fairness audit to evaluate whether prediction outcomes differ systematically across sensitive student groups. Applied to a real-world dataset of 6,607 students with 19 features, our best model (XGBoost) achieves R² = 0.725, RMSE = 1.971. For 96% of sampled low-performing students, FairCF successfully generated valid counterfactual explanations by recommending realistic changes to actionable behavioural features such as attendance and tutoring sessions. The demographic fairness audit reveals the largest Demographic Parity Gap (DPG = 0.100) for Family Income, while Gender and School Type exhibit near-perfect parity. We connect explainable AI with educational equity by providing fairness-aware decision support for educators and students.Cite

Cloud robotics is emerging as a scalable and cost-effective alternative to traditional robotic systems by leveraging the power of cloud computing. However, its dependence on network connectivity exposes it to a wide range of cybersecurity threats. This research introduces a hybrid machine learning framework designed specifically to detect multiclass cyber threats in cloud robotics systems. Using the CIC IoT-DIAD 2024 dataset, the study evaluates the performance of various machine learning and deep learning models, including Random Forest, XGBoost, Logistic Regression, CNN, and Feedforward Neural Networks. Among these, the Random Forest classifier achieved the highest accuracy of 99.32%, outperforming other state-of-the-art models. A hybrid model combining CNN and Random Forest was also tested, offering improved robustness. The proposed approach not only strengthens the security of cloud robotic systems but also contributes to real-time threat detection, scalability, and autonomous system integrity.Cite