Analysis of Density between Classes for Mitigating Class Imbalance Problem for Activity Recognition
Sayeda Shamma Alia, Sozo Inoue,
Class Imbalance problem is unavoidable problem for activity recognition using mobile sensors in real life scenario when we have very less amount of data for some activity classes. It can affect the accuracy of the algorithms for classification. In this paper we measure the impact of class density in the accuracy of classification in imbalance cases. It is important for understanding the problem better that can help in finding better solution for classification in this scenario. Our initial experiment shows that- class imbalance affects the performance of classifier negatively and the higher the value of density (lower deviation), the better the performance of the classifier becomes.