What Sheeba demonstrated

Built an end-to-end turbine yield predictor using leakage-free sensor inputs and a genuine time-based split. Compared a LinearRegression baseline against a tuned RandomForest, then evaluated the final model on held-out later-period data with MAE, RMSE, R2, and residual analysis. The model card clearly states intended use, excluded downstream emissions features, limitations, and a monitoring plan.

• Leakage-Free Feature Selection
• Time-Based Model Validation
• Honest Held-Out Evaluation
• Residual Diagnostics