Downsampling. This graph shows the effect of downsampling on the classifier’s detection rate. The detection rate exponentially increases until it starts stagnating once it exceeds 20 datapoints per sample. In the bottom right corner of this figure, we show an example of the confusion matrix for 20 datapoints. Confusion matrix legend: TP: True Positive, FN: False Negative, FP: False Positive, TN: True Negative, TPR: True Positive Ratio, FPR:False Positive Ratio.