Supervised Learning

Supervised Learning is the most commonly used type of ML. The model learns from pairs of (input, correct output).

Classification vs Regression

Classification - Predicting Categories

1from sklearn.datasets import make_classification
2from sklearn.linear_model import LogisticRegression
3
4# Generate Safari data
5X, y = make_classification(n_samples=1000, n_features=4, n_classes=3)
6# y: 0 = Lion, 1 = Elephant, 2 = Cheetah
7
8# Logistic Regression for classification
9model = LogisticRegression()
10model.fit(X_train, y_train)
11
12# Predicting class
13prediction = model.predict([[180, 200, 4, 0.8]])  # Lion?
14probabilities = model.predict_proba([[180, 200, 4, 0.8]])

Regression - Predicting Continuous Values

1from sklearn.linear_model import LinearRegression
2
3# Predicting population based on reserve area
4X = [[100], [200], [300], [400]]  # area in km²
5y = [150, 280, 420, 550]           # population
6
7model = LinearRegression()
8model.fit(X, y)
9
10# Prediction
11new_area = [[250]]
12predicted_population = model.predict(new_area)
13print(f"Predicted population: {predicted_population[0]:.0f}")

Evaluation Metrics

Classification Metrics

1from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
2from sklearn.metrics import confusion_matrix, classification_report
3
4y_true = [0, 1, 1, 0, 1, 1, 0, 0]
5y_pred = [0, 1, 0, 0, 1, 1, 1, 0]
6
7# Accuracy - % of correct predictions
8print(f"Accuracy: {accuracy_score(y_true, y_pred):.2%}")
9
10# Precision - how many of the predicted positives were correct
11print(f"Precision: {precision_score(y_true, y_pred):.2%}")
12
13# Recall - how many of the actual positives did we detect
14print(f"Recall: {recall_score(y_true, y_pred):.2%}")
15
16# F1-score - harmonic mean of precision and recall
17print(f"F1: {f1_score(y_true, y_pred):.2%}")
18
19# Confusion Matrix
20print(confusion_matrix(y_true, y_pred))
21# [[TN, FP],
22#  [FN, TP]]
23
24# Full report
25print(classification_report(y_true, y_pred))

Regression Metrics

1from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
2import numpy as np
3
4y_true = [100, 200, 300, 400]
5y_pred = [110, 190, 310, 380]
6
7# MSE - Mean Squared Error
8mse = mean_squared_error(y_true, y_pred)
9print(f"MSE: {mse:.2f}")
10
11# RMSE - Root Mean Squared Error
12rmse = np.sqrt(mse)
13print(f"RMSE: {rmse:.2f}")
14
15# MAE - Mean Absolute Error
16mae = mean_absolute_error(y_true, y_pred)
17print(f"MAE: {mae:.2f}")
18
19# R² - coefficient of determination (0-1, closer to 1 is better)
20r2 = r2_score(y_true, y_pred)
21print(f"R²: {r2:.4f}")

Cross-validation

1from sklearn.model_selection import cross_val_score, KFold
2
3model = RandomForestClassifier()
4
5# 5-fold cross-validation
6scores = cross_val_score(model, X, y, cv=5)
7print(f"CV scores: {scores}")
8print(f"Mean: {scores.mean():.2%} (+/- {scores.std() * 2:.2%})")
9
10# Custom KFold
11kfold = KFold(n_splits=5, shuffle=True, random_state=42)
12for train_idx, val_idx in kfold.split(X):
13    X_train, X_val = X[train_idx], X[val_idx]
14    y_train, y_val = y[train_idx], y[val_idx]
15    # Train and evaluate...