InsureLead

Predicting Customer Responses to Insurance Offers Using ML

Project Duration: Jul 15, 2024 - Aug 1, 2024

---

🌟 Introduction

The objective is to predict which customers will respond positively to a vehicle insurance offer. This project is part of a binary classification challenge which was hosted on Kaggle. Submissions were evaluated using Area Under the ROC Curve (AUC).

---

🥉 Top Approach

Explore full implementation here: 🔗 PS4E7 - Stacking Boosters with ANN

• 📊 Data Integration & Inspection

  • Combined official training dataset with original insurance dataset for feature enrichment.

• 🛠️ Preprocessing Pipelines

  • Utilized Scikit-learn pipelines and transformers with encoders: StandardScaler, PowerTransformer, OneHotEncoder, OrdinalEncoder.

• 🔍 Feature Engineering & Selection

  • Applied mutual information filtering to retain informative features.

• 🧰 Modeling with Ensembles

  • Trained and validated XGBoost, CatBoost, LightGBM classifiers using Stratified K-Fold CV.
  • Hyperparameter tuning with Optuna and visual exploration tools.

• 🏋️ Submission Strategy

  • Ensemble predictions via model averaging on test data.

---

📊 Results / Outcomes

• ✅ Public Leaderboard Scores: ranging from 0.50060 to 0.89727

• 🏁 Best Private Score: *0.89690

• 🥇 Rank Achieved: Ranked 70 / 2425 participants and 2234 teams as a solo participant

---

🔗 References

• 📁 Kaggle Competition: Binary Classification of Insurance Cross Selling
  

• 📂 Original Dataset: Health Insurance Cross Sell Prediction Data

---

🛠️ Tech Stack

• Language: Python 🐍

• Libraries:

  • pandas, polars, numpy for data handling

  • matplotlib, seaborn for EDA and plotting

  • scikit-learn, xgboost, catboost, lightgbm for modeling

  • optuna for hyperparameter tuning

• Tools:

  • Jupyter Notebook / Kaggle Notebooks for experimentation

  • Custom pipelines and scoring functions for AUC optimization

---