Online feature selection for contextual time series data ( Extended abstract )

We propose a simple and efficient method for online feature selection from time series data. Our method is based on calculating characteristics of the different features and calculating similarity values for feature pairs using Gaussian kernels. Our motivation has been to design a method that can be used to select the most relevant context features for activity recognition. Namely, traditional feature selection methods have been designed for offline use and thus are not applicable in our setting. The efficiency of our method is evaluated using toy data and real context data, gathered using a 3D accelerometer.