Machine Learning Pipeline & Metrics

Training Phase

• Goal: Minimize training error while avoiding overfitting.
• Model learns patterns from the data.
• Use training loss to guide optimization.

Testing (or Validation) Phase

• Goal: Measure generalization error.
• Evaluate on unseen data to check performance.
• Metrics here tell if the model is truly good.

Linear Regression

• Predicts continuous values.
• Example: Predicting house prices.
• Multiple Linear Regression → multiple predictors.

Logistic Regression

• Classification task, despite the name.
• Predicts probabilities for categories.
• Example: Spam vs. Not Spam.

For Regression (Continuous Target)

• R² → Higher is better.
• MSE / RMSE / RSS → Lower is better.
• p-values (for statistical significance).

For Classification (Including Logistic Regression)

• Accuracy → Higher is better.
• Precision, Recall, F1-score.
• ROC-AUC.