Degree

Bachelor of Science (Computer Science)

Department

Department of Computer Science

School

School of Mathematics and Computer Science (SMCS)

Advisor

Ms. Tasbiha Fatima, Lecturer, Department of Computer Science

Co-Advisor

Malik Qamar Hayat - Manager - Habib Metro Bank

Keywords

Footfall Forecasting, Banking Analytics, LightGBM, Predictive Analytics, Queueing Theory, Business Intelligence, Role-Based Access Control

Abstract

This project presents an end-to-end branch intelligence platform developed for Habib Metro Bank, integrating operational analytics, predictive footfall forecasting, and an AI-powered natural language query interface. The system consists of a role-aware React dashboard, a FastAPI backend with PostgreSQL and Redis, and a machine learning pipeline trained on more than one million banking transaction records. Three distinct dashboard views were designed for branch managers, regional supervisors, and executives, providing secure and role-specific insights. The forecasting engine evaluates nine machine learning models using time-based validation, with LightGBM achieving the best performance (R² = 0.918, MAE = 2.53 customers/hour, RMSE = 7.25). The model incorporates 55 engineered features, including temporal patterns, holiday effects, Ramadan, Eid, Friday Juma, and salary-week indicators. Forecast outputs are further translated into teller deployment recommendations using an M/M/c queueing model, enabling data-driven staffing decisions. The platform also includes simulated real-time updates through Redis and WebSockets, role-based access control, multi-factor authentication, and an AI assistant for querying branch performance metrics. Experimental results demonstrate the platform’s effectiveness in forecasting customer footfall and supporting operational decision-making in retail banking environments.

Tools and Technologies Used

Python, FastAPI, React.js, JavaScript, PostgreSQL, Redis, LightGBM, XGBoost, Scikit-learn, Optuna, SQLAlchemy, JWT Authentication, WebSockets, Docker, Docker Compose, Tailwind CSS, Recharts, Vite, Nginx, Ollama, Large Language Models (LLMs), REST APIs, Redis Pub/Sub, PostgreSQL, MFA/TOTP, Git, Machine Learning, SHAP, Queueing Theory (M/M/c Erlang-C)

Methodology

The project followed a full-stack, data-driven development methodology comprising data engineering, machine learning, backend development, and frontend implementation. Historical banking transaction data (1M+ records) was preprocessed and transformed into hourly footfall datasets. Feature engineering was performed using temporal, seasonal, holiday, and contextual variables, resulting in 55 predictive features. Nine machine learning models were trained and evaluated using a chronological train-validation-test split and walk-forward cross-validation, with LightGBM selected as the final model based on performance metrics. The forecasting outputs were integrated with an M/M/c queueing model to generate teller deployment recommendations. A FastAPI backend with PostgreSQL and Redis was developed to serve analytics, forecasts, and AI-powered insights through secure REST and WebSocket APIs. The frontend was implemented using React and Tailwind CSS, providing role-based dashboards for branch managers, regional supervisors, and executives. The complete solution was containerized using Docker and evaluated through performance, accuracy, security, and usability testing.

Document Type

Restricted Access

Submission Type

BSCS Final Year Project

Download

Share

COinS