Performance Analysis of Feature-Based Automated Measurement of Mouse Social Behavioral

Abstract

Automated social behavior analysis in the  mammalian animal has become an increasingly popular  and attractive alternative to traditional manual human  annotation with the advancement of machine learning  and video tracking system for automatic detection. In  this work, we study a framework of how different  features perform on the different classifiers to analyze  automatic mice behavior. We conducted experiments on  the Caltech Resident-Intruder Mouse (CRIM13) dataset,  which provides two types of features: trajectory features  and spatio-temporal features. With this feature, we train  AdaBoost and Random Decision Forest (TreeBagger)  classifiers to classify different mouse behaviors to show  which features perform best on which classifier. The  experimental result shows that the trajectory features  are more informative and provide better accuracy than  the widely used spatio-temporal features, and AdaBoost  classifier shows better performance than the TreeBagger  on these features.