Improvement of Linear Discriminant Analysis by Applying The Ensemble Method

Linear Discriminant Analysis (LDA) is one of the learning algorithms for the binary problems. One of the drawbacks of LDA is the degradation of its performance when it is difficult to approximate the boundary between two classes with a linear function, i.e. the boundary is non-linear. In this research, we apply Error-Correcting Output Codes (ECOC) to LDA. ECOC is well-known as one of the ensemble methods. Further, we introduce the weighted voting method to reduce the influence of poor hypotheses on the classification accuracy. The hypotheses are weighted according to their classification accuracy. Moreover, although LDA was originally designed for applying to only binary classification problems, it can be extended to the multi-class problems by introducing ECOC.