IMI Big Data and AI Competition (Sept 2025 - Feb 2026)
This repository contains the data analysis pipeline, machine learning models, and strategic presentation developed for the IMI Big Data and AI Competition. The core objective was to build a robust detection model that minimizes financial loss while maintaining a frictionless user experience for legitimate customers.
Note: Due to data privacy and competition confidentiality, the original dataset of 61,000+ records has been removed. A mock_data.zip archive has been provided to demonstrate the schema and pipeline functionality without compromising proprietary information.
Fraud detection isn't just about catching bad actors; it's about reducing false positives that alienate good customers. By moving beyond basic rule-based systems, this project delivered:
- 22% Optimization in Fraud Detection: Successfully applied supervised learning and statistical analysis to drastically improve detection rates over the baseline model.
- Actionable Stakeholder Dashboards: Translated complex model outputs into clear visual dashboards to drive strategic decision-making and product strategy for non-technical stakeholders.
The project required a full-stack data science approach, focusing on translating analytical findings into business value.
- Data Engineering: Cleaned, transformed, and engineered features from a massive, multi-table transactional dataset.
- Predictive Modeling: Implemented and tuned supervised machine learning models to identify high-risk transaction patterns.
- Data Visualization & Strategy: Built comprehensive dashboards to visualize risk distribution, ensuring the technical findings directly informed business strategy.
IMI_Fraud_Detection_Pipeline.ipynb: The complete Python data analysis and machine learning pipeline (configured to run with the provided mock data).IMI_Fraud_Detection_Deck.pdf: The strategic presentation deck outlining the business case, methodology, and final product recommendations.mock_data.zip: A synthetic dataset illustrating the data schema required to run the pipeline.