Abstract
In pattern recognition, there are two widely used methods, MT (Mahalanobis Taguchi) and machine learning. The advantage of the MT method is that it makes it possible to detect unknown abnormal patterns. However, the discrimination accuracy is relatively lower than machine learning. In addition, there is a constraint on the variables. Alternatively, machine learning has higher discrimination accuracy than the MT method; however, discrimination of unknown abnormal patterns may be difficult.
The purpose of this study is to propose a method that utilizes the strengths of both methods and eliminates the drawbacks of them. By applying AdaBoost to the MT method, the proposed method is evaluated in terms of the following. First, discrimination accuracy comparable to that obtained using machine learning is realized. Second, it is possible to accurately distinguish unknown abnormal patterns, which may pose a challenge when using machine learning. Third, the variable constraint condition in the MT method can be solved. Through analytics of breast cancer data and a created simulation data based on it, we demonstrate the superiority of the proposed method over conventional methods when the data is likely to include unknown abnormal patterns.
