A Deep Learning-Powered Tool for Early Detection of Student Depression with SHAP Interpretability

Vijayaragavan, Aarthi (2025) A Deep Learning-Powered Tool for Early Detection of Student Depression with SHAP Interpretability. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (2MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Depression among university students continues to be a major global concern, affecting not only academic performance, but also personal well-being and future potential. Conventional methods for identifying depression, such as questionnaires or clinical assessments, are often time-consuming, subjective, and not practical for early detection on a large scale. This research proposes a deep learning-based approach using a Multilayer Perceptron (MLP) model to predict depression from structured academic and lifestyle data, including CGPA, sleep duration, financial stress, and study satisfaction. To overcome the common “black box” issue of deep learning, the model integrates SHAP (SHapley Additive Explanations) to make predictions more transparent and interpretable by highlighting the factors most strongly influencing the outcome.

Experimental results show that the SHAP-enhanced MLP achieved an accuracy of 84.6% and an AUC score of 0.918, performing better than traditional machine learning models such as Logistic Regression and Random Forest. The training process required 38 seconds in a standard CPU environment, while the inference time per prediction averaged 12 milliseconds, making the model suitable for real-time applications. The global explanations via SHAP computation took 6.7 seconds and through feature importance analysis the input variables were reduced to only 10 from 77 without loss of predictive performance.

A user-friendly Streamlit application was developed to provide real-time prediction and visual explanations, making the model both practical and ethical for real-world use. In general, this work demonstrates how explainable deep learning can support universities and mental health professionals in identifying at-risk students early and taking meaningful preventive action.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Yaqoob, Abid UNSPECIFIED
Uncontrolled Keywords:	Student Depression Prediction; Deep Learning; Multilayer Perceptron (MLP); SHAP Interpretability; Real-Time AI Tool; Streamlit Interface
Subjects:	L Education > LB Theory and practice of education > LB2300 Higher Education Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning R Medicine > RA Public aspects of medicine > RA790 Mental Health
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	09 Sep 2026 11:11
Last Modified:	09 Sep 2026 11:11
URI:	https://norma.ncirl.ie/id/eprint/9927

Actions (login required)

View Item