3.2 Comparison of ML methods
A classifier algorithm to compare ten ML methods was done. Fig. 3 shows a graphical representation of the relationship between % Accuracy and Log Loss values generated by the classifier algorithm, for each of the tests performed with the ten ML methods selected previously. In this figure, the marker color is associated with the name of the ML method, and the marker shape with the concentration range of the test.
Moreover, from these results, the most efficient methods were selected. Four ML methods with the lower value of Log Loss (closer to zero) and a higher value of % Accuracy (ML methods with % Accuracy values close to 100%) were selected. The selected methods were: RFC, SVM, NuSVC, and GBC. Next, these four ML methods were evaluated by a ROC analysis, obtaining as a result that the best ML model for classification is the one carried out by the RFC algorithm, demonstrating that RFC has a better probability of sensing.
Another reason for which RFC was chosen is that when there is binary discrimination, the precision parameter is decisive. Since the RFC model was the one with the highest precision value in the different tests, it was selected as the optimum model. This means that the number of false positives using this model is low.