Machine Learning Evaluation Metrics
Common ML evaluation metrics: Accuracy, Precision, Recall, F1 Score, ROC-AUC, MAE, MSE, and R-squared with formulas and best-use scenarios.
Evaluation metrics for machine-learning models with type, formula, best use and range: accuracy, precision, recall, F1, ROC-AUC, MAE, MSE and R-squared, extended with specificity, log loss, RMSE, Matthews correlation, IoU, perplexity, BLEU, nDCG and Cohen's kappa.
Use it when
- You need the formula and interpretation of a metric.
- You are choosing a metric for imbalanced classification, ranking or generation.
Watch out for
- A single metric hides trade-offs; report several and choose by the cost of different errors.
- Metrics such as BLEU correlate imperfectly with human judgment of quality.
- ROC-AUC can range from 0 to 1; 0.5 is a random ranking, not the lower bound. Report prevalence and the decision threshold alongside threshold-specific metrics.
The data
Metrics
21 rows
| Name | Type | Formula | Best for | Range |
|---|---|---|---|---|
| Accuracy | Classification | (TP + TN) / (TP + TN + FP + FN) | Balanced datasets | [0, 1], higher is better |
| Precision | Classification | TP / (TP + FP) | Minimizing false positives | [0, 1], higher is better |
| Recall | Classification | TP / (TP + FN) | Minimizing false negatives | [0, 1], higher is better |
| F1 Score | Classification | 2 * (Precision * Recall) / (Precision + Recall) | Imbalanced datasets | [0, 1], higher is better |
| ROC-AUC | Classification | Area under ROC curve | Ranking quality | [0, 1]; 0.5 is the random-ranking baseline, and lower values are possible |
| MAE | Regression | mean(|y_true - y_pred|) | Average error magnitude | [0, inf), lower is better |
| MSE | Regression | mean((y_true - y_pred)^2) | Penalizing large errors | [0, inf), lower is better |
| R-squared | Regression | 1 - (SS_res / SS_tot) | Proportion of variance explained | (-inf, 1], higher is better |
| Specificity | Classification | TN / (TN + FP) | Measuring how well negatives are identified | [0, 1], higher is better |
| Log Loss | Classification | -(1/N) * sum(y log p + (1 - y) log(1 - p)) | Evaluating predicted probabilities | [0, infinity), lower is better |
| Matthews Correlation Coefficient | Classification | (TP*TN - FP*FN) / sqrt((TP+FP)(TP+FN)(TN+FP)(TN+FN)) | Imbalanced binary classification | [-1, 1], higher is better |
| RMSE | Regression | sqrt(mean((y - y_hat)^2)) | Regression where large errors should weigh more, in the units of the target | [0, infinity), lower is better |
| Intersection over Union | Detection and segmentation | |A intersect B| / |A union B| | Overlap between predicted and true regions | [0, 1], higher is better |
| Perplexity | Language modeling | exp(-(1/N) * sum(log p(w_i | context))) | Comparing language models on the same text | [1, infinity), lower is better |
| BLEU | Machine translation | Brevity penalty * geometric mean of n-gram precisions | Comparing translations with references | [0, 1], higher is better |
| nDCG | Ranking | DCG / ideal DCG | Search and recommendation quality with graded relevance | [0, 1], higher is better |
| Cohen's Kappa | Agreement | (p_o - p_e) / (1 - p_e) | Agreement between annotators beyond chance | [-1, 1], higher is better |
| Average Precision | Ranking / binary classification | sum((recall_n - recall_previous) * precision_n) | Summarizing the precision-recall tradeoff when positives are rare | [0, 1], higher is better; prevalence affects the baseline |
| Brier Score (binary) | Probability prediction | mean((p_positive - y_binary)^2) | Evaluating probabilistic predictions alongside calibration plots | [0, 1], lower is better |
| Balanced Accuracy | Classification | mean(recall_per_class) | Giving each class equal weight in the recall average | [0, 1], higher is better |
| Pinball Loss | Quantile regression | mean(max(q * error, (q - 1) * error)); error = y_true - y_pred | Evaluating a specified conditional quantile | [0, inf), lower is better |
Fields
The shape of this dataset, read from the data itself: every column, its type and a real value. Agents can use this to plan a query without fetching the whole document.
| Section | Field | Type | Example |
|---|---|---|---|
| metrics | name | string | Accuracy |
| metrics | type | string | Classification |
| metrics | formula | string | (TP + TN) / (TP + TN + FP + FN) |
| metrics | best_for | string | Balanced datasets |
| metrics | range | string | [0, 1], higher is better |
Fetch the same bytes
The static files contain the same reference rows as the API, with no API rate limit. Live metadata, such as view counts, may differ. Use the API for content negotiation; use the static files for a cacheable snapshot.
curl "https://yjtoon.yunusemrevurgun.com/api/dataset/ml-evaluation-metrics?format=toon"
const res = await fetch( "https://yjtoon.yunusemrevurgun.com/static-data/dataset/ml-evaluation-metrics.toon" ); const toon = await res.text();
Rate limit: 120 requests per minute per IP, no key and no signup. API reference →
Sources and editorial notes
Original editorial reference notes by YJTOON, informed by the linked primary documentation. External specifications retain their own licenses.
Editorial review: . Source documents keep their own licenses.