Beta-Liouville and Inverted Beta-Liouville Based Predictive Models for Occupancy Detection using Small Training Data

Jiaxun Guo, Manar Amayri, Wentao Fan, Nizar Bouguila

Abstract: In this paper, we propose two predictive models based on Beta-Liouville (BL) and inverted Beta-Liouville (IBL) mixture models. The choice of the BL and IBL mixture models is motivated by their flexibility. The proposed predictive models are dedicated to classification tasks where the training datasets are non-Gaussian and small which is generally the case in real-life scenarios. A principled variational approach is proposed to learn the proposed models. Extensive experimental results based on both synthetic data and a real application that concerns occupancy detection in smart buildings prove that our predictive framework achieves promising results especially with extremely small training data sets.