Machine Learning Specialization Skills Covered

Supervised learning uses labeled examples to learn a mapping from inputs to outputs. Linear regression predicts continuous values, while logistic regression handles classification. Candidates often confuse logistic regression with ordinary regression because of the name; the output and loss function matter.

Gradient descent updates model parameters to reduce a cost or loss function. If the learning rate is too large, training may diverge; if it is too small, progress is slow. Feature scaling can make optimization behave more predictably.

Feature engineering can improve separability or capture nonlinear patterns, but polynomial features can increase variance. Regularization discourages overly complex models; too little regularization may overfit, while too much can underfit.

Use training data for fitting, validation or cross-validation data for model selection, and a test set for final evaluation. Repeatedly tuning against the test set leaks information and makes performance estimates too optimistic.

Accuracy can mislead when classes are imbalanced. Precision is useful when false positives are costly; recall matters when false negatives are costly; F1 balances precision and recall. Threshold selection changes the tradeoff.

Neural networks combine layers, activation functions, forward propagation, loss functions, and backpropagation. For this specialization, focus on the intuition and implementation basics rather than deep research-level architectures.

Decision trees split data into regions but can overfit when allowed to grow too complex. Random forests and boosting reduce or manage tree weaknesses in different ways. The practical question is whether the model generalizes, not whether the tree perfectly fits training data.

Clustering groups unlabeled examples, anomaly detection finds unusual examples, and recommender systems estimate preferences. These are not ordinary classification tasks because the target signal and evaluation workflow differ.