Insurance company

This project demonstrated how predictive modeling—particularly using logistic regression, ensemble methods, and behavioral flags—can effectively forecast insurance plan purchases. These insights enable data-driven decision-making for product optimization, improving customer satisfaction and business performance.
Modeling Strategy:
1. Base Hypothesis:
Final purchase often mirrors the last quote reviewed → 94% accuracy in training, 53.79% in testing (due to quote history truncation).

2. Enhanced Prediction Model:
Step 1: Logistic Regression + PCA to identify customers likely to change their plan (68% accuracy)

Step 2: For those likely to change, used:

Multinomial Logistic Regression (to predict each of the 7 components separately)

Random Forest Classifier (to predict entire 7-digit vector jointly — achieved 64.85% accuracy)

3. Post-processing:
Replaced unlikely insurance plans (chosen <5% of the time) with most common alternatives based on similar profiles