AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
One-vs-Rest (OvR)
Strategy for decomposing multi-class classification into multiple binary classifiers, where each classifier distinguishes one class against all the others.
One-vs-One (OvO)
Approach that trains a binary classifier for each possible pair of classes, requiring n(n-1)/2 classifiers for n classes.
Multi-class confusion matrix
An NxN table where N is the number of classes, displaying the model's correct and incorrect predictions to evaluate its classification performance.
F1 macro score
Unweighted average of the F1 scores of each class, treating all classes equally regardless of their frequency in the dataset.
F1 micro score
Metric that calculates the F1 score globally by counting the total true positives, false negatives, and false positives across all classes.
F1 weighted score
Weighted average of the F1 scores per class, where each class contributes proportionally to its number of samples in the dataset.
Logits
Unnormalized raw scores produced by the last layer of a neural network before applying the softmax function for multi-class classification.
Multi-class cross-entropy
Loss function that measures the divergence between the predicted probability distribution and the actual distribution of labels in a multi-class context.
Stratified sampling
Sampling technique that preserves the proportion of classes in training and test subsets to avoid distribution bias.
Class imbalance
Situation where classes are not represented equally in the dataset, which can bias learning toward majority classes.
Random undersampling
Undersampling method that randomly reduces the number of samples from majority classes to balance the class distribution.
Random oversampling
Oversampling approach that randomly duplicates samples from minority classes to achieve balance with majority classes.
Multi-class precision
Metric calculated as the average of precisions per class, where the precision of a class is the ratio of true positives to the total positive predictions.
Multi-class recall
Metric representing the average of recalls per class, measuring the model's ability to identify all relevant samples for each class.
Multi-class ROC curve
Extension of the ROC curve for multi-class classification, generally calculated using a One-vs-Rest or macro-average approach.
Multi-class AUC
Area under the ROC curve for multi-class problems, calculated as the average of binary AUCs according to the One-vs-Rest approach.
Multi-class decision trees
Rule-based algorithms that recursively partition the feature space to create leaves corresponding to the different classes.
Multi-class logistic regression
Extension of logistic regression using the softmax function to model the probability of belonging to each of the K possible classes.
Multi-class SVM
Support vector machines adapted for multi-class classification, typically implemented via One-vs-Rest or One-vs-One strategies.
Multi-class KNN
K-nearest neighbors algorithm that classifies a new observation according to the majority class among its k nearest neighbors.
Multi-class Naïve Bayes
Probabilistic classifier based on Bayes' theorem with conditional independence assumption, naturally extended to multi-class problems.
Multi-class Random Forest
Ensemble method combining multiple decision trees trained on bootstrap subsamples, with classification by majority vote.