Chemical reaction optimization to disease diagnosis by optimizing hyper-planes classifiers

Early diagnosis of diseases can save and leads to survival. There are several diagnoses techniques which mostly consist of classification and optimization parts. Although these techniques have their specific advantages, they have their significant disadvantages such as sensitivity to the number of features (symptoms) and need to features selection, challenge to detect non-integrated regions of one class and high complexity of their progresses. In this paper to fill up the disadvantages, a novel classification is proposed to disease diagnosis by different numbers of hyper-planes classifiers (HPC) that divides medical data into adequate regions based on assigning binary codes to each region. The HPC can find useful relationships between the symptoms of the diseases by tagging each region with the suitable class label. To optimize the HPC’s coefficients and improve disease diagnosis, chemical reaction optimization (CRO) is adapted based on four reactions on HPC’s coefficients, which are coded as molecular structures. Different numbers of HPCs are performed, and their experimental results are compared together. The interesting point of the results is disease diagnosis error 0.000% by five hyper-planes for test data of all investigated medical data set. Also, the best-obtained results of the CRO-HPC are compared with the best outputs of more than 50 methods of disease diagnosis from the previous state-of-the-art literature. This comparison shows that CRO-HPC’s diagnosis errors can compete with the majority of the other diagnostic methods.