7005SCN_Individual_Research_Project

A Comparative Analysis of Machine Learning and Deep Learning Models for Predicting Student Depression with a Simple Prototype Early Warning System

Author

Venkata Siva Sai Prasad Mannava

Project Overview

This dissertation develops and evaluates Machine Learning (ML) and Deep Learning (DL) models for predicting student depression using a publicly available Student Depression Dataset. The study compares the predictive performance of multiple models after applying comprehensive data preprocessing and hyperparameter optimisation. A simple desktop-based prototype early warning system is also developed using the best-performing model to demonstrate the practical application of student depression prediction.

Research Aim

To comparatively evaluate the performance of Machine Learning and Deep Learning models for predicting student depression using hyperparameter tuning and multiple evaluation metrics, and to develop a simple prototype early warning system for academic demonstration.

Research Objectives

Preprocess the Student Depression Dataset by handling missing values, encoding categorical variables, scaling numerical features, and balancing the classes. Develop Machine Learning and Deep Learning models for student depression prediction. Optimise model performance using hyperparameter tuning. Compare all developed models using multiple evaluation metrics. Develop a simple prototype early warning system using the best-performing model. Dataset

Dataset Name

Student Depression Dataset

Source Target Variable

Depression

Classification Type

Binary Classification

Technologies Used

Python Jupyter Notebook Pandas NumPy Matplotlib Seaborn Scikit-learn LightGBM XGBoost CatBoost TensorFlow/Keras Imbalanced-learn (SMOTE) Tkinter

Data Preprocessing

The following preprocessing steps were performed:

Data loading Missing value analysis Duplicate record checking Label Encoding Feature Standardisation (StandardScaler) Train-Test Split SMOTE for class balancing Machine Learning Models LightGBM XGBoost CatBoost Deep Learning Models Artificial Neural Network (ANN) Multi-Layer Perceptron (MLP) Hyperparameter Optimisation

Hyperparameter tuning was performed for all developed models to improve prediction performance and model generalisation.

Evaluation Metrics

The developed models were evaluated using:

Accuracy Precision Recall F1-Score (Primary Evaluation Metric) ROC-AUC Confusion Matrix

Experimental Results

Model Accuracy Precision Recall F1-Score ROC-AUC LightGBM 86.78% 0.8640 0.8730 0.8685 0.9423 XGBoost 86.84% 0.8630 0.8757 0.8693 0.9416 CatBoost 87.12% 0.8627 0.8828 0.8726 0.9424 ANN 85.95% 0.8513 0.8711 0.8611 0.9326 MLP 85.63% 0.8656 0.8436 0.8544 0.9330 Best Performing Model

CatBoost achieved the highest predictive performance among all developed models.

Performance

Accuracy: 87.12% Precision: 0.8627 Recall: 0.8828 F1-Score: 0.8726 ROC-AUC: 0.9424

Therefore, CatBoost was selected as the final prediction model for the prototype early warning system.

Prototype

A desktop-based prototype was developed using Python Tkinter.

Features User-friendly interface Student information input Automatic preprocessing Depression prediction Displays: Depression Detected No Depression Detected

The prototype demonstrates the practical application of the developed predictive model for academic purposes.

mannavav/7005SCN_Individual_Research_Project

About

Resources

Stars

Watchers

Forks

Releases

Languages