Telecom Churn Prediction

📈 Data Analysis & Feature Engineering

• Conducted a correlation matrix analysis to explore relationships between variables such as usage frequency, complaints, and customer status

• Applied Variance Inflation Factor (VIF) analysis to detect multicollinearity (i.e., highly correlated predictors), and refined the dataset by removing or transforming problematic variables to enhance model interpretability

🤖 Predictive Modeling & Optimization

Tested multiple classification algorithms including:

• Logistic Regression

• K-Nearest Neighbors (KNN)

• Quadratic Discriminant Analysis (QDA)

• Naive Bayes

Refined the logistic regression model by selecting significant predictors (e.g., call failures, complaints, subscription length), improving model accuracy from 87.9% to 89.9%

🔎 Model Evaluation & Threshold Tuning

Evaluated models using standard metrics:

• Accuracy

• Confusion Matrix (True Positives, True Negatives, False Positives, False Negatives)

• Area Under the Receiver Operating Characteristic Curve (AUC-ROC) — achieved a high AUC score of 0.93, indicating strong sensitivity and model performance

• Optimized the classification threshold from the default 0.5 to 0.51, reducing error rate and improving predictive precision

  

🔁 Validation Strategy

• Employed 10-Fold Cross-Validation, a model evaluation technique that partitions the dataset into 10 parts to ensure robustness and minimize overfitting