Performance evaluation of a multi-stage classification for cow behavior

Decision tree (DT) algorithm is a simple and effective method for classification of cow behavior. In order to evaluate the performance of this classifier, Receiver-Operating Characteristic (ROC) curve is used. However, this classification often consists of multiple stages of simpler classifiers with multi-level decision thresholds. This paper proposes to use a parallel of multiple thresholds in order to maximize one of statistical measures of performance (i.e. sensitivity, specificity, and precision), after the evaluation of the performance of the multi-stage system using a series of ROC and thresholds is done. The proposed method is then applied to the existing experimental data provided in literature in order to highlight its advantages.