Welcome to Engage2Value Analytics

Gain insights and predict purchase values instantly.

🔢
--

Total Features

🎯
purchaseValue

Target Variable

✓
Active

Model Status

⚡ Key Insights: Global Behavioral Drivers

Deep Dive into the top two predictive features: pageViews and totalHits

Marginal Effect of pageViews

Observe the non-linear plateau effect as page views increase.

Loading PDP...

Anomaly Detection in totalHits

Isolation Forests highlight anomalous high-hit sessions (Red X).

Loading Outliers...

🛡️ Engage2Value: From Clicks to Conversions

Predicting Customer Purchase Value from Multi-Session Digital Behavior

Problem Statement

The goal of this project is to predict a customer’s purchase value based on their multi-session behavior across digital touchpoints.

The target variable purchaseValue is extremely right-skewed, highly zero-inflated, and influenced by complex non-linear interactions.

Key Challenges

  • Heavy class imbalance (majority zero purchase)
  • Extremely right-skewed target (skew > 50)
  • High-cardinality categorical features
  • Risk of data leakage via user/session IDs
  • Multicollinearity among strong behavioral features

Feature Engineering & Modeling

Two parallel datasets were maintained: a Leakage-free dataset and an ID-inclusive dataset for controlled comparison.

Feature Stacking added major signal through K-Means Clustering and a Binary Purchase Classifier (LightGBM).

The final solution utilizes an Ensemble Learning approach achieving a robust ~0.70 public R² without relying on IDs.

🧭 How to Use This Application

🔍 Advanced Data Exploration

Navigate to the Data Exploration tab to access enterprise-grade ML interpretability:

  • Global SHAP: See the top 10 features driving the model globally.
  • Isolation Forests: Detect and highlight anomalous user behavior in red.
  • Partial Dependence Plots: View the non-linear marginal effect of features.
  • Bivariate Analysis: Analyze the intersection of two features via cross-tabulation heatmaps and color-coded scatter plots.

🤖 Purchase Prediction

Navigate to the Prediction Form to simulate customer behavior:

  • Manually input session data like browser, hits, and page views.
  • Use the Randomize Inputs button to quickly test the model logic.
  • See the expected purchase value instantly generated by our stacked ensemble model.

Built for the Machine Learning Practice Project.
Predict, explore, and convert with confidence.

Advanced Data Exploration

Select a feature to generate statistical inferences, distributions, and SHAP importance.

🌍 Global Feature Importance (SHAP)

Top 10 features driving the predictive model globally across all segments.

Loading global SHAP calculations...

Purchase Prediction

Enter feature values or randomize to predict the expected purchase value.