An Optimized Feature Selection and Deep Learning Framework for Kawasaki Disease Detection in Clinical Data Mining

Abstract

Kawasaki disease (KD) is a form of acute vasculitis, which has been accompanied with coronary aneurysm of the artery, coronary artery dilation, arrhythmia and other deadly heart diseases. Even nowadays, its etiology is not clear enough; the molecular mechanism and the associated issues of KD should be investigated. Over the past few years deep learning has replaced the state-of-the-art in a number of tasks in medicine.  The KNN (K-Nearest Neighbour) algorithm was the first algorithm that was used in imputing missing data. The optimization algorithm the Adapted Bee Colony with Firefly (ABCFF) is then used to do the feature selection i.e. selection of the best feature out of training data. In addition, the Firefly algorithm generates a new location where the new location of the scout bee stage will be replaced. Its functionality is tested against some benchmark functions.  The f-score values that they consider as the best features selection criteria are the values of fitness applied in the present work. The parameter-optimized convolutional neural network (CNN) is used in the classification, and particle swarm optimization is used to figure out the best parameters of the CNN. The existing method can improve clinical data mining. The paper reported that a feature representation using matrices and feature extraction using CNN can be applied on mine clinical data which is left unfinished to facilitate the medical decision making.