Learning from life-logging data by hybrid HMM: a case study on active states prediction

In this paper, we have proposed employing a hybrid classifier-hidden Markov model (HMM) as a supervised learning approach to recognize daily active states from sequential life-logging data collected from wearable sensors. We generate synthetic data from real dataset to cope with noise and incompleteness for training purpose and, in conjunction with HMM, propose using a multiobjective genetic programming (MOGP) classifier in comparison of the support vector machine (SVM) with variant kernels. We demonstrate that the system with either algorithm works effectively to recognize personal active states regarding medical reference. We also illustrate that MOGP yields generally better results than SVM without requiring an ad hoc kernel.