Performance evaluation in activity classification: factors to consider

Abstract

After building the model to recognize activities from sensor data, it is essential to investigate the effectiveness of the model. The evaluation of the performance for machine learning methods can be performed using some evaluation matrices. This chapter properly explains the evaluation matrices namely accuracy, precision, recall, F1 score, balance classification rate, confusion matrix, and so on. Graphical performance measures namely ROC curve, cumulative gains, and lift charts have been explained too. This chapter also represents some essential concepts related to precision and recall trade-off, and accuracy as a performance measure. The contents of this chapter will be useful not only for human activity recognition, but also for other classification-related researches.