Machine learning for medical diagnosis: A neural network classifier optimized via the directed bee colony optimization algorithm

Abstract A fast and accurate method for diagnosis is required in the medical domain. The existing techniques use classification methodology but are not accurate and fast enough or viable for medical diagnosis. In this chapter, we present a high performance artificial neural network (ANN) optimized via the directed bee colony (DBC) algorithm. The methodological analysis is utilized to diagnose cancer, diabetes, and heart disease. The performance analysis is done on three principal variants for diagnosis: classification accuracy, uniqueness of solution, and running time. The relatively low performance of earlier techniques remains a critical barrier for the successful transaction of these techniques for medical diagnosis. The ANN is trained with a newly developed DBC algorithm, which mimics the group food search and decision-making strategy of bees to avoid the premature convergence to local minima. The coalition of bees incorporates the parallelism in an optimal solution search, which reduces the running time. The algorithm results in a unique solution after every run, thereby increasing the medical viability of the proposed method. Extensive comparison with other evolutionary algorithms needs to be done and is tabulated, which highlights its upshots in terms of uniqueness, classification accuracy, and running time. A comprehensive comparison from different meta-analytic studies is done on 16 other algorithms. In two different criterion methods, DBC has been ranked second and first. In terms of uniqueness of the answer, DBC has been ranked first both times. Also, the running time is approximately 101% and 21% more than DBC by GA and PSO.