About This Project
Forecasted telecom network usage using Poisson, Negative Binomial, Ridge, and Lasso regression models, addressing overdispersion and multicollinearity. Reduced RMSE by 34% through optimized modeling and feature engineering. Classified customer complaints using Logistic Regression with SMOTE oversampling, improving AUC from 0.8206 to 0.924 and accuracy to 87.5%. Applied regularization techniques (Lasso, Ridge, Elastic Net) and validated models with cross-validation. Visualized model results and insights in Tableau.
Key Features
- Poisson & Negative Binomial regression for count data with overdispersion
- Ridge, Lasso, Elastic Net regularization for multicollinearity
- Logistic Regression + SMOTE: AUC improved from 0.82 → 0.924
- 34% RMSE reduction through feature engineering and model tuning
- Stratified cross-validation for robust performance estimates
- Tableau dashboards for network usage and complaint insights
