Machine Learning Roadmap

What is ML?

Programs that learn patterns from data instead of being explicitly programmed.

Traditional: rules + data → answers
ML:          data + answers → rules (a model)

Core concepts

• Supervised vs Unsupervised Learning — the two main paradigms
• Bias-Variance Tradeoff — underfitting vs overfitting
• Loss Functions — what models optimize
• Cross-Validation — robust evaluation
• Evaluation Metrics — accuracy, precision, recall, F1, AUC
• Hyperparameter Tuning — finding the best model settings
• Regularization — preventing overfitting

Supervised Learning — Regression

• Linear Regression — the simplest model, the foundation
• Polynomial Regression — fitting curves

Supervised Learning — Classification

• Logistic Regression — linear classification, probability outputs
• Decision Trees — interpretable, rule-based
• Random Forests — ensemble of trees, robust
• Gradient Boosting — XGBoost, LightGBM — often wins competitions
• Support Vector Machines — maximum margin classifiers
• K-Nearest Neighbors — classify by similar examples
• Naive Bayes — probabilistic, fast, good for text

Unsupervised Learning

• K-Means Clustering — partition data into k groups
• PCA — dimensionality reduction, find principal directions
• Anomaly Detection — find unusual data points

Learning order

1. Linear Regression → understand loss, gradient descent, evaluation
2. Logistic Regression → extend to classification
3. Decision Trees → understand non-linear models
4. Random Forests + Gradient Boosting → practical workhorse models
5. Evaluation + Cross-Validation → proper methodology
6. Unsupervised → clustering, PCA when needed

Links

• Data Fundamentals Roadmap — data preparation feeds into modeling
• Deep Learning Roadmap — when classical ML isn’t enough