Hybrid feature reduction and selection for enhanced classification of high dimensional medical data

The objective of this study is to investigate the effectiveness of hybrid feature reduction and selection for the efficient classification of Dementia. The availability of an effective method that is more objective than human readers is needed to produce more reliable dementia diagnostic procedures. The proposed scheme consists of several steps including dimensionality reduction, followed by feature selection and classification by Support vector machine. This research paper proposes a embedded feature selection method which improves the classification performance of the support vector machine. The large volumes of features extracted from brain Magnetic Resonance Images and neuropsychological tests may lead to less efficient classification. Hence the hybrid approach which is trained with multiple biomarkers effectively reduced the high dimensional data set and facilitated accurate classification when compared with conventional feature reduction and feature selection techniques. Non-linear Kernel functions of SVM that are varied and compared with reduced data set provided nearly 97% accuracy and 96.5% sensitivity. Features selected by Gain ratio filter improved the performance of the Support Vector Machine classifier when compared with Information gain and Correlation filters.