Improving machine learning accuracy in diagnosing diseases using feature selection based on the fruit- fly algorithm

Data mining methods can be used to classify people as sick or healthy and to diagnose diseases. One of the important challenges in diagnosing the disease using learning methods such as multilayer artificial neural network is the optimal selection of its parameters that can be solved to some extent with meta-heuristic methods, but these techniques are not without problems because in these methods meta-heuristic algorithm To a certain extent, it can only reduce the error, and to reduce the machine learning error, in addition to the optimal selection of learning parameters, it is necessary to consider and select the appropriate features. The proposed method has two layers and in the first layer the selection of weight and bias of the multilayer neural network is done using the fruit fly algorithm and in the input layer, the feature selection phase is done using the binary version of the fruit fly optimization algorithm to learn to Do it only on the important features and reduce the problem space and increase the learning speed and accuracy. The results of tests on several disease datasets in the UCI database show that the accuracy, sensitivity, and diagnosis of the proposed method are 98.36%, 98.12%, and 98.08%, respectively. The proposed method is more accurate in diagnosing diabetes than the PSO, FA, SHO, and HHO algorithms.