Naurose Farhan

Academic Projects

CSE370: Database Systems

GameVault DBMS

GameVault is a full-stack web application designed for discovering, buying and renting digital video games. The application is categorized as a digital distribution platform for PC gaming. The website connects a frontend user interface to a backend server and a MySQL database. The project integrates core e-commerce features, enabling users to browse a diverse game catalog and choose between two purchasing options: buy or rent. For rentals, users can choose from three different duration options. Additionally, the platform promotes community engagement through a dedicated newsfeed that delivers the latest gaming articles. The standout feature of GameVault’s marketplace is the renting system of games. Traditional PC gaming storefronts, such as Steam and Epic Games, offer only the option to buy games at full price. GameVault offers not only the option to buy but also the ability to rent games for 7, 14, or 30 days, with rental prices adjusted based on the game’s full price. This system enables users to experience a game for the specified days before deciding whether to make a permanent purchase to keep the game in their library. The web application is built using HTML for structure, CSS3 for styling, and JavaScript for logic. The Node.js server handles user requests and processes them based on the data stored in the MySQL database.

Technology Stack: HTML, CSS, Express, Node.js, JavaScript, MySQL, JWT

Project Report: Documentation / Github: Project-GameVault

CSE422: Artificial Intelligence

Project: Predicting the Severity of Depression using Machine Learning Models

The objective of this project is to analyze surveyed data on depression and apply machine learning models that can predict the severity of depression levels based on various parameters such as demographic, academic and psychological factors. In order to classify, a full pipeline including data preprocessing, learning models, and performance evaluation is implemented. By applying supervised and unsupervised learning techniques, this project attempts to identify important patterns in student mental health data and compare the performance of different machine learning algorithms.

Model Training & Testing:

• Logistic Regression (with class balancing to reduce bias from class imbalance)
• Random Forest Classifier (implemented tree-based ensemble; scaling not required)
• Gaussian Naive Bayes
• K-Nearest Neighbors (KNN) (implemented required scaling)
• Neural Network (MLPClassifier) with hidden layers (64, 32, 16) using ReLU activation and Adam optimizer. Early stopping enabled to reduce overfitting.

Model Selection/Comparison Analysis:

Multiple evaluation metrics were used to compare the models. Accuracy was shown as an overall score, while macro-averaged precision, recall, and F1-score were reported to account for class imbalance. Confusion matrices were plotted for each model to visualize per-class performance. Multiclass ROC curves were generated using the One-vs-Rest approach and AUC scores were computed for comparing classifier separability. The accuracy comparison shows that Logistic Regression achieved the highest accuracy. Random Forest and Neural Network achieved moderate performance. Lastly, Naive Bayes and KNN showed comparatively lower accuracy.

Project Report: Documentation / Github: AI Project