Evaluation

Model evaluation is a crucial step in machine learning, ensuring that models perform well on unseen data. This section provides an overview of different evaluation techniques available in EvoML for classification, regression, and time series models.

Classification Evaluation

Classification models predict categorical outcomes. EvoML provides several evaluation methods to assess classification performance:

1. Metrics: Includes accuracy, precision, recall, F1-score, and more.
2. Confusion Matrix: Visual representation of predicted vs. actual classes.
3. Cumulative Gains Chart: Measures model effectiveness in identifying positive instances.
4. Cumulative Lift Chart: Evaluates improvement over random selection.
5. Precision-Recall Curve: Shows trade-offs between precision and recall.
6. Probability Calibration Graph: Assesses how well predicted probabilities align with true probabilities.
7. Receiver Operating Characteristic Curve: Evaluates classification performance at different thresholds.

Regression Evaluation

Regression models predict continuous values. EvoML offers various techniques to measure model accuracy:

1. Metrics: Includes Mean Absolute Error, Mean Squared Error, and R² score.
2. Prediction Scatter: Visualizes predicted vs. actual values.
3. Residual Scatter: Highlights prediction errors.
4. Box Plot: Summarizes model residual distribution.

Time Series Evaluation

Time series models predict values based on temporal sequences. EvoML provides:

1. Line Plot: Displays actual vs. predicted values over time.

Understanding evaluation metrics and visualizations helps in interpreting model performance and making informed improvements. Refer to individual evaluation guides for deeper insights into each method.