Weak false label learning model for sensor data recognition

Real world behavior recognitions tend to suffer from incomplete data because sensors are not perfect. Although machine learning algorithms are successfully applied to recognitions, they do not work well in multi-valued output functions because true and false label in same input collide in learning process. In this paper, we propose a noble algorithm which lessens multi-valued function´s problem by weakening false labels. It also includes virtual samples and output normalization to compensate for the balance between true and false labels.