The use of machine learning for building a classifier in signal processing for motion sensing presents unique challenges. This paper proposes a novel method that effectively addresses the combination of skewed data sets and optimization requirements. By utilizing a customized loss function and a product of probability models, our approach achieves a fully automated and efficient machine learning process. Additionally, our resulting probability models offer reduced complexity, making them ideal for embedded applications. Our method offers a promising solution for motion sensing applications that require accurate and efficient classification.