Nahian Arnob

This project analyses credit card transaction data to identify and understand fraudulent behaviour using exploratory data analysis and machine learning techniques. The analysis explores transaction patterns across fraud and non-fraud cases, including class imbalance, transaction amounts, and time-based trends. Multiple visualizations are used to highlight differences in spending behavior, and baseline classification models are developed to evaluate fraud detection performance using precision-focused metrics. The project serves as a prototype fraud detection system and demonstrates practical approaches used in real-world financial risk analytics.

Millions of people use urban subway transportation systems every day, and they are an important part of modern cities. Subway Transit authorities can make better use of their resources, change the frequency of service, cut down on passenger wait times, and make the whole system more efficient if they can accurately predict how many people will use a stop or route. This paper presents the design, implementation, and deployment of an end-to-end data-driven transit demand forecasting system for the New York City subway network. The system integrates 3 real-world public datasets: MTA Subway Hourly Ridership data (51 million rows, 2022–2024), MTA GTFS Static Network feeds, and NYC Permitted Events data for 2026. The dataset was filtered to focus on subway ridership for the 2024 calendar year, resulting in a clean dataset of 10 million records and a final processed feature set of 1,396,290 station-hour observations across 16 engineered features. 5 machine learning and deep learning models were trained and evaluated: XGBoost, Random Forest, Long Short-Term Memory (LSTM), Prophet, and Temporal Fusion Transformer (TFT). Model performance was assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). Amongst all, XGBoost achieved the best overall performance with an MAE of 349.04 riders per station per hour and an RMSE of 923.64, and was selected as the deployment model. The trained model was deployed through a two-tier architecture consisting of a FastAPI RESTful backend serving real-time predictions and a professional Streamlit web application featuring an interactive NYC subway map, forecast controls, confidence interval visualizations, and automated decision support alerts for transit planners. The complete system was implemented and deployed locally within a single development session, demonstrating the practical feasibility of machine learning-powered transit forecasting as a tool for smarter urban transit planning and operational decision-making.