Introduction to Machine Learning
Concepts • Algorithms • Evaluation
A concise, original study guide
Original educational guide Page 1
Table of Contents
1. Machine Learning in Simple Terms
2. Main Learning Types
3. Data Preparation
4. Training, Validation and Test Sets
5. Regression
6. Classification
7. Decision Trees and Ensembles
8. Clustering
9. Neural Networks
10. Model Improvement Checklist
Original educational guide Page 2
1. Machine Learning in Simple Terms
Machine learning is a field of computing in which algorithms learn patterns from data and use those
patterns to make predictions or decisions.
A typical workflow is: collect data → clean data → represent features → train a model → validate it →
evaluate it → deploy and monitor it.
2. Main Learning Types
Supervised learning learns from examples with known target values. Classification predicts categories;
regression predicts numerical values.
Unsupervised learning finds structure without labeled targets, such as clusters or lower-dimensional
representations.
Semi-supervised learning combines a smaller labeled dataset with a larger unlabeled dataset.
3. Data Preparation
Real-world data can contain missing values, duplicates, inconsistent formats and outliers. Cleaning should
be performed carefully so that useful information is not accidentally removed.
Feature scaling is useful for algorithms that depend on distances or gradient-based optimization. Common
approaches include standardization and min-max scaling.
4. Training, Validation and Test Sets
Training data is used to fit model parameters. Validation data helps choose settings such as model
complexity. Test data provides a final estimate of performance.
Data leakage occurs when information unavailable at prediction time accidentally enters the training
process. Leakage can produce unrealistically strong results.
5. Regression
Linear regression models a numerical target using a weighted combination of features. A simple form is y
= b0 + b1x.
Evaluation measures include mean absolute error (MAE), mean squared error (MSE) and root mean
squared error (RMSE).
6. Classification
Classification predicts a discrete class. Logistic regression, decision trees, k-nearest neighbors and
support vector machines are common approaches.
Useful metrics include accuracy, precision, recall and F1-score. For imbalanced classes, accuracy alone
may be misleading.
7. Decision Trees and Ensembles
Original educational guide Page 3
A decision tree repeatedly splits data using feature conditions. It is easy to interpret but can overfit.
Ensemble methods combine multiple models. Random forests average many randomized trees, while
boosting builds models sequentially to correct earlier errors.
8. Clustering
K-means clustering assigns observations to a chosen number of clusters by minimizing within-cluster
squared distance to cluster centers.
The number of clusters can be explored using domain knowledge, the elbow method or silhouette
analysis.
9. Neural Networks
A neural network contains layers of connected units. Each unit applies a weighted sum followed by an
activation function.
Training adjusts weights to reduce a loss function, commonly using gradient-based optimization and
backpropagation.
10. Model Improvement Checklist
Start with a simple baseline. Verify the data split. Choose metrics that match the problem. Compare
models fairly. Tune hyperparameters using validation or cross-validation. Inspect errors instead of looking
only at one score.
Finally, consider deployment constraints such as latency, memory, interpretability, data drift and
monitoring.
Original educational guide Page 4