Supervised Learning - Prediction Week 3 Challenge
Ques1. In attached file
Ques2.
L1 Regularization (Lasso Regression)
Adds the absolute value of coefficients to the loss function.
Effect:
-
Can shrink some coefficients to zero, leading to sparse models.
-
Helps in feature selection.
L2 Regularization (Ridge Regression)
Adds the square of coefficients to the loss function.
Effect:
-
Shrinks coefficients towards zero, but not exactly to zero.
-
Helps when many small/medium features are important.
Ques3.
MSE (Mean Squared Error)
MAE (Mean Absolute Error)
Huber Loss
Log-Cosh Loss
Quantile Loss
Ques4.
Learning Rate (α) controls the step size in Gradient Descent.
Too high → overshoot minima → divergence.
Too low → slow convergence → stuck in local minima.
Ideal α helps the model converge fast and accurately.
Ques6.
Multiple Coefficient of Determination (R^2): Measures how well the independent variables explain the variability in the dependent variable. Always increases (or stays the same) when more variables are added—even if they are not significant. Good for initial understanding but may overfit if many predictors are used.
Adjusted Coefficient of Determination (Adjusted R^2): Modified version of R² that adjusts for the number of predictors. Only increases if new predictors improve the model significantly. Better for comparing models with different numbers of predictors.
Ques5.
| Metric | Description |
|---|---|
| R-Squared (R^2) | Proportion of variance explained |
| Adjusted R^2 | R^2 adjusted for number of predictors |
| MSE/RMSE | Error magnitude (lower is better) |
| MAE | Average absolute error |
| Visualizations | Residual plot, actual vs predicted |