E-Commerce Shopper Behavior Analysis

Project Type: Academic (UCF) | Date: Oct. 2025 – Dec. 2025 | Team Size: Individual

--- ## Business Problem **Objective:** Predict purchase intent from user session data to optimize: - **Marketing Spend:** Target high-intent users with ads - **Inventory Management:** Anticipate demand spikes - **User Experience:** Personalize content for conversion **Challenge:** Severe class imbalance (only 15.5% of sessions result in purchases) --- ## Dataset **UCI Machine Learning Repository: Online Shoppers Purchasing Intention Dataset** | Attribute | Description | |-----------|-------------| | **Samples** | 12,330 sessions | | **Features** | 17 (10 numerical, 7 categorical) | | **Target** | Revenue (Binary: Purchase / No Purchase) | | **Class Distribution** | 84.5% No Purchase, 15.5% Purchase | | **Time Period** | 1 year of e-commerce activity | **Key Features:** - **Behavioral:** Page views (Administrative, Informational, Product) - **Engagement:** Time on site, bounce rates, exit rates - **Temporal:** Month, weekend vs. weekday - **User Context:** Visitor type (New/Returning), browser, OS, region - **Google Analytics:** Page values (revenue attribution) --- ## Methodology ### 1. Exploratory Data Analysis (EDA) **Class Imbalance Investigation:** ```python import pandas as pd import matplotlib.pyplot as plt # Class distribution revenue_counts = df['Revenue'].value_counts() print(f"No Purchase: {revenue_counts[False]} ({revenue_counts[False]/len(df)*100:.1f}%)") print(f"Purchase: {revenue_counts[True]} ({revenue_counts[True]/len(df)*100:.1f}%)") # Visualization plt.figure(figsize=(8, 6)) revenue_counts.plot(kind='bar', color=['#ff6b6b', '#4ecdc4']) plt.title('Class Imbalance in Revenue') plt.xlabel('Purchase Decision') plt.ylabel('Count') plt.xticks(rotation=0) plt.show() ``` **Result:** 84.5% / 15.5% split → SMOTE required --- ### 2. Feature Engineering **Log Transformations:** Heavy right-skew in numerical features required log transformation: ```python from sklearn.preprocessing import FunctionTransformer import numpy as np log_features = ['ProductRelated', 'ProductRelated_Duration', 'PageValues'] log_transformer = FunctionTransformer(np.log1p, validate=True) df[log_features] = log_transformer.transform(df[log_features]) ``` **Before/After:** Reduced skewness from 5.2 → 0.8 (closer to normality) **One-Hot Encoding:** ```python from sklearn.preprocessing import OneHotEncoder categorical_features = ['Month', 'VisitorType', 'Browser', 'Region'] encoder = OneHotEncoder(sparse=False, drop='first') # Avoid multicollinearity encoded = encoder.fit_transform(df[categorical_features]) feature_names = encoder.get_feature_names_out(categorical_features) df_encoded = pd.DataFrame(encoded, columns=feature_names, index=df.index) df = pd.concat([df.drop(categorical_features, axis=1), df_encoded], axis=1) ``` --- ### 3. Unsupervised Learning: K-Means Clustering **Goal:** Discover latent user segments before supervised modeling **Elbow Method for K:** ```python from sklearn.cluster import KMeans from sklearn.metrics import silhouette_score inertias = [] silhouette_scores = [] for k in range(2, 11): kmeans = KMeans(n_clusters=k, random_state=42, n_init=10) kmeans.fit(X_scaled) inertias.append(kmeans.inertia_) silhouette_scores.append(silhouette_score(X_scaled, kmeans.labels_)) # Optimal K = 5 (elbow + silhouette peak) ``` **5 Discovered Segments:** 1. **Bargain Hunters:** Low PageValues, high bounce rate, weekend traffic 2. **Window Shoppers:** High page views, low conversion, returning visitors 3. **Intent Buyers:** High PageValues, low exit rate, product-focused 4. **Impulse Buyers:** Short sessions, high conversion rate 5. **Researchers:** Long duration, high informational pages, low conversion **Feature Contribution:** Added cluster assignments as categorical feature → +2.3% AUC improvement --- ### 4. Statistical Testing **ANOVA (Numerical Features):** ```python from scipy.stats import f_oneway for feature in numerical_features: purchasers = df[df['Revenue'] == True][feature] non_purchasers = df[df['Revenue'] == False][feature] f_stat, p_value = f_oneway(purchasers, non_purchasers) if p_value < 0.001: print(f"{feature}: F={f_stat:.2f}, p={p_value:.2e} *** SIGNIFICANT") ``` **Significant Features (p < 0.001):** - PageValues: F=1245.32 - ExitRates: F=892.14 - ProductRelated_Duration: F=456.78 **Chi-Square (Categorical Features):** ```python from scipy.stats import chi2_contingency for feature in categorical_features: contingency_table = pd.crosstab(df[feature], df['Revenue']) chi2, p_value, dof, expected = chi2_contingency(contingency_table) if p_value < 0.05: print(f"{feature}: χ²={chi2:.2f}, p={p_value:.2e} * SIGNIFICANT") ``` **Significant Categorical Features:** - Month (χ² = 234.56, p < 0.001) - VisitorType (χ² = 189.23, p < 0.001) --- ### 5. Handling Class Imbalance: SMOTE **Synthetic Minority Over-sampling Technique:** ```python from imblearn.over_sampling import SMOTE smote = SMOTE(sampling_strategy='auto', random_state=42, k_neighbors=5) X_train_resampled, y_train_resampled = smote.fit_resample(X_train, y_train) print(f"Before SMOTE: {Counter(y_train)}") print(f"After SMOTE: {Counter(y_train_resampled)}") ``` **Result:** - Before: {False: 8,405, True: 1,539} - After: {False: 8,405, True: 8,405} → Balanced 50/50 --- ### 6. Model Comparison Pipeline **Scikit-learn Pipeline Architecture:** ```python from sklearn.pipeline import Pipeline from sklearn.preprocessing import StandardScaler from sklearn.model_selection import GridSearchCV models = { 'KNN': KNeighborsClassifier(), 'Logistic Regression': LogisticRegression(max_iter=1000), 'SVM': SVC(probability=True), 'Random Forest': RandomForestClassifier(random_state=42) } for name, model in models.items(): pipeline = Pipeline([ ('scaler', StandardScaler()), ('classifier', model) ]) pipeline.fit(X_train_resampled, y_train_resampled) y_pred = pipeline.predict(X_test) print(f"\n{name} Results:") print(classification_report(y_test, y_pred)) ``` --- ## Results ### Model Performance Comparison | Model | Accuracy | Precision | Recall | F1-Score | AUC-ROC | Training Time | |-------|----------|-----------|--------|----------|---------|---------------| | KNN | 0.85 | 0.72 | 0.68 | 0.70 | 0.84 | 2.1s | | Logistic Regression | 0.88 | 0.79 | 0.76 | 0.77 | 0.88 | 0.8s | | SVM | 0.87 | 0.77 | 0.74 | 0.75 | 0.87 | 45.3s | | **Random Forest** | **0.91** | **0.88** | **0.86** | **0.91** | **0.93** | **12.4s** | ### Winner: Random Forest **Hyperparameters (GridSearchCV):** ```python param_grid = { 'classifier__n_estimators': [100, 200, 300], 'classifier__max_depth': [10, 20, None], 'classifier__min_samples_split': [2, 5, 10] } grid_search = GridSearchCV(pipeline, param_grid, cv=5, scoring='f1', n_jobs=-1) grid_search.fit(X_train_resampled, y_train_resampled) print(f"Best params: {grid_search.best_params_}") # Best: n_estimators=300, max_depth=20, min_samples_split=2 ``` **Final Test Set Performance:** - **AUC-ROC:** 0.93 - **F1-Score:** 0.91 - **Precision:** 0.88 (88% of predicted purchases are correct) - **Recall:** 0.86 (86% of actual purchases are caught) --- ### Feature Importance Analysis **Top 10 Predictive Features:** | Feature | Importance | Interpretation | |---------|------------|----------------| | **PageValues** | 0.342 | Revenue attribution from Google Analytics | | **ExitRates** | 0.186 | Higher exit = lower purchase intent | | **ProductRelated_Duration** | 0.124 | Time spent on product pages | | **Month_Nov** | 0.089 | Black Friday effect | | **BounceRates** | 0.067 | Single-page sessions | | **Administrative_Duration** | 0.052 | Time on account/checkout pages | | **VisitorType_Returning** | 0.041 | Loyalty indicator | | **Weekend** | 0.033 | Shopping behavior differences | | **Cluster_3** | 0.028 | Intent Buyer segment | | **SpecialDay** | 0.021 | Proximity to holidays | **Key Insight:** PageValues alone accounts for 34.2% of model decisions --- ## Business Recommendations 1. **Targeting Strategy:** - Users with PageValues > 50: 78% purchase probability → Aggressive retargeting - Exit Rate > 0.05: Offer exit-intent popups with discounts 2. **Seasonality:** - November (Black Friday): 2.3x baseline conversion → Scale inventory by 130% - Weekend traffic: Lower conversion but higher order value → Focus on premium products 3. **User Segmentation:** - **Intent Buyers (Cluster 3):** Streamline checkout, reduce friction - **Window Shoppers (Cluster 2):** Email nurturing campaigns, social proof 4. **Real-Time Scoring:** - Deploy model as API: Predict purchase intent within first 30 seconds - Use predictions to personalize product recommendations and discounts --- ## Challenges & Learnings | Challenge | Solution | Learning | |-----------|----------|----------| | **Severe Class Imbalance** | SMOTE + class weights | Always visualize class distribution first | | **High Feature Correlation** | VIF analysis, drop correlated pairs | ExitRates & BounceRates had VIF > 10 | | **Overfitting** | Cross-validation, max_depth tuning | Validation curve showed depth=20 optimal | | **Interpretability** | Feature importance + SHAP values | Stakeholders need "why", not just "what" | --- ## Technologies Used Python • Scikit-learn • Pandas • NumPy • Matplotlib • Seaborn • SMOTE • Jupyter --- ## Links & Resources - **💻 [GitHub Repository](https://github.com/jacksonSmall/shopper-behavior-analysis)** - **📧 [Questions?](mailto:jacksonSmall@ucf.edu)** --- ## Key Takeaways