An evaluation of supervised, novelty-based and hybrid approaches to fall detection using Silmee accelerometer data