Fintech Needs-Based Recommendation System

🎯 Project Mission & My Role

As part of the FinTech course at Politecnico di Milano, our team was tasked with a critical, real-world challenge: building a financial recommendation engine that is not only accurate but also transparent, ethical, and fully compliant with regulations like MiFID II. As the project owner and a key contributor, I guided our team's strategy to develop a "needs-first" recommendation system. Instead of a black-box model, we engineered an Explainable AI (XAI) pipeline using interpretable decision trees. My focus was on balancing high performance with absolute regulatory compliance and ensuring our final solution was both intelligent and trustworthy. This project sits at the intersection of my passions: data science, financial technology, and building systems that create real, responsible value.

---

🤝 The Team

This project's success was a result of a talented and collaborative team effort.

• Mehdi Ghiasipour
• Reza Ilzadi Jahromi
• Hirdesh Kumar
• Suraj Manickam Thirumalai Raja
• Christos Partasidis

---

📌 Overview

This project builds a needs-first recommendation system for retail banking clients using interpretable machine learning. Instead of starting from financial products, we identify individual client needs — Income vs Accumulation — and match suitable products from a catalog using a transparent, explainable model.

---

🎯 Objectives

• Detect client investment needs (Income / Accumulation) based on demographic and financial data.
• Use interpretable decision trees to ensure explainability.
• Recommend a suitable product that adheres to client-specific risk tolerances.

---

🧩 Files

• Fintech_Final_Project_Estimating_Needs_V1_3.ipynb
  Main implementation in Jupyter Notebook. Sections include:

  • Data ingestion and EDA
  • Feature engineering
  • Model training and validation (Decision Trees)
  • Explainability and Partial Dependence Plots (PDPs)
  • Next Best Action (NBA) logic for product recommendation

• Fintech_Final_Project_Team_15.pdf
  Accompanying presentation deck summarizing the project:

  • WHAT / WHY / HOW / RESULTS
  • Visuals and metrics
  • Regulatory context and business motivation

---

🧠 How It Works

1. Data

• Rows: 5000 anonymized retail bank clients
• Features:
  • Age, Gender, FamilyMembers, FinancialEducation, RiskPropensity, Income, Wealth
• Targets:
  • Binary flags: IncomeInvestment, AccumulationInvestment

2. Feature Engineering

• Gender x Age → life-stage modeling
• log(Income / Wealth) → liquidity vs capital insight
• Log-transform + MinMax scaling for skewed data

3. Model

• Two separate Decision Trees (depth=5) for the two need targets
• 5-fold cross-validation
• F1 score used as primary evaluation metric

4. Recommendation Logic (NBA)

• Predict each client’s needs
• Filter product catalog by need type
• Enforce risk suitability: ProductRisk ≤ ClientRiskPropensity
• Recommend the highest-risk suitable product

---

📊 Results

Target	Baseline F1	F1 w/ Engineering	Δ F1	Precision	Recall
IncomeInvestment	0.689	0.729	+0.04	0.77	0.79
AccumulationInvestment	0.680	0.818	+0.14	0.70	0.72

• Recommendation Coverage
  • IncomeInvestment: 94% (227/241 matched)
  • AccumulationInvestment: 75% (357/475 matched)
  • 0 risk cap violations (MiFID II compliant)

---

🛠️ How to Run

1. Install dependencies

   pip install pandas numpy scikit-learn matplotlib seaborn