Feature analysis applying clustering and optimisation methods to Mahalanobis-Taguchi method

While data analysis is important in various corporate activities, it is often the case that a company's data analysis is not well-conducted. There are two main reasons for this: the lack of teacher data and the increasingly complicated nature of the data to be analysed, which makes it difficult to judge the appropriate analysis unit/group and to select the appropriate items to be used for the analysis. In response, we propose a data analysis approach that combines a clustering and a stochastic optimisation model with the Mahalanobis-Taguchi method, making it possible to automatically determine the group of data to be analysed and the items of data to be used, and to extract features from the data. The proposed approach enables data analysis with a single correct label and eliminates tasks that require higher-level skills (such as feature selection). The effectiveness of the proposed method is verified using recorded TV data.