Integrated feature preprocessing for classification based on neural incremental attribute learning

Incremental Attribute Learning (IAL) is a feasible machine learning strategy for solving high-dimensional pattern classification problems. It gradually trains features one by one, which is quite different from those conventional machine learning approaches where features are trained in one batch. Preprocessing, such as feature selection, feature ordering and feature extraction, has been verified as useful steps for improving classification performance by previous IAL studies. However, in the previous research, these preprocessing approaches were individually employed and they have not been applied for training simultaneously. Therefore, it is still unknown whether the classification results can be further improved by these different preprocess approaches when they are used at the same time. This study integrates different feature preprocessing steps for IAL, where feature extraction, feature selection and feature ordering are simultaneously employed. Experimental results indicate that such an integrated preprocessing approach is applicable for pattern classification performance improvement. Moreover, statistical significance testing also verified that such an integrated preprocessing approach is more suitable for the datasets with high-dimensional inputs.