Classification performance on scenarios given at three different difficulty levels.

The k-nearest neighbors algorithm was tested on three two-class classification scenarios at three difficulty levels. The scenario difficulty was defined as the similarity between two gas classes. The classification model was trained under 10-fold cross-validation procedure with 10 repetitions, and the best value of the k parameter was estimated along possible values 3, 5, 7 and 9 for each classification model. The accuracy in prediction of class labels was used to score the models. The model complexity, expressed in value of parameters k, is observed to increase with greater scenario difficulty. The first model provides a perfect performance with a 100% rate of classification, while the last model displays poor accuracy with a classification rate of 0.74 on the test set.