Improved linear discriminant analysis for feature selection in heart disease prediction

In this research work, a novel heart disease prediction framework is developed with four major phases, viz. 'pre-processing, feature extraction, feature selection and heart disease prediction phase'. Initially, the collected raw data is pre-processed via data cleaning approach. Then, from the pre-processed data, the most relevant features like the 'higher order statistical features including central tendency (arithmetic means, median, mode, standard deviation, geometric mean, harmonic mean, interquartile mean, midrange, midhinge, trimean and winsorised mean), degree of dispersion (range, average absolute deviation, coefficient of variation, relative mean difference, Gini coefficient, mean absolute difference, entropy) and symmetrical uncertainty' is extracted. Subsequently, the features are selected using the I-LDA approach. Then, a new hybrid classifier is constructed in the heart disease prediction phase, which is trained with the selected features.