Neural Networks Using Fuzzy Perceptrons in Medical Datasets

Abstract

Fuzzy logic is a generalization of traditional logic that works with values more familiar to natural language, allowing problems to be solved that would be very difficult to solve using binary logic. It goes hand in hand with other areas of AI, such as evolutionary computing and neural networks. In the latter, it is thought that the potential of fuzzy logic can be exploited even further, which is why the fuzzy perceptron is proposed. This proposal, a combination of the perceptron with fuzzy logic, consists of applying an activation function with a fuzzy system that determines, through rules, the output of the artificial neuron based on the weighted sum of its inputs and weights. Medical datasets have a wide variety of data compositions, such as different data volumes, data types, organization, etc., which is useful for training neural networks. Therefore, experiments are carried out using various clinical datasets to train neural networks with the proposed activation function and compare them with training a traditional neural network with the aim of achieving improved accuracy in the classification of clinical data.