Degree

Bachelor of Science (Computer Science)

Department

Department of Computer Science

School

School of Mathematics and Computer Science (SMCS)

Advisor

Tasbiha Fatima, Lecturer, Department of Computer Science, Institute of Business Administration, Karachi

Keywords

Ai/ML, FYP, Webapp

Abstract

This project introduces an AI-based credit scoring system developed in collaboration with Meezan Bank, aimed at enhancing the accuracy and efficiency of credit risk assessment through modern machine learning techniques. The core objective is to leverage artificial intelligence to analyze a range of financial and behavioral attributes in order to generate reliable credit scores. Trained on financial data collected by the Pakistani Govt, labeled HIES 2018/2019, the system adopts a student-teacher distillation paradigm, where the student model learns from credit scores generated by a hardcoded rule-based formula, a neural network, and a gradient boosting algorithm. This design enables the student model to generalize complex patterns effectively while leveraging the complementary strengths of each scoring approach. The current implementation successfully demonstrates the feasibility of AI in streamlining credit scoring, offering promising implications for more inclusive and data-driven financial decision-making within the banking sector.

Tools and Technologies Used

Python, Tensor Flow, Django, Scikit Learn, Postgres, Tailwind, Google Colab, Vscode, Kaggle

Methodology

The traditional methods of credit-scoring rely on outdated techniques of score calculation which result in numerous disadvantages for both the banks and its clients. The process for credit scoring typically involves manually performing redundant calculations on data relating to each client which becomes quite a hefty process on banks’ side due to the overhead of employing several employees just for the score calculation. Similarly, this process also has disadvantages on clients’ side. One of these disadvantages is the clients’ receiving unfair scores as this methodology of credit scoring is based on a shallow formula-based approach which usually fails to capture the vast number of parameters and the weight of each parameter. An automated AI-based approach would solve both of these problems as the scoring would be a data-driven process based on thousands of records and dozens of features, which for the most part would be performed by a computer.

This project proposes the development of an AI-based credit scoring system which would counter several problems faced using the traditional system. By utilizing machine learning algorithms, the final product would harness the power of computers to deal with large volumes of data and perform repetitive tasks robustly. The application would take vast amount of client data such as transaction history, financial records and personal data to regress to a dynamically calculated accurate credit score. The platform will enable lenders to make better-informed decisions, minimize default risks, and improve access to credit for undeserved populations. It will also ensure compliance with relevant financial regulations and maintain robust data security standards.

Following an agile methodology framework we aimed to have constant meetings with the industry for our app while also having major changes and deliverables per sprint with a constant feedback loop accordingly. Though our approach changed severely across the schedule due to changing requirements from the industry, we implemented them in due time via our agile framework and went from a low level AI model to an ensemble student teacher distilled model with semi supervised learning.

Document Type

Restricted Access

Submission Type

BSCS Final Year Project

Download

Share

COinS