Pediatric HSCT survival rates detection through the application of different ANN model optimizers, BSA, VSA, and GOA

Abstract

Abstract Machine learning (ML) possesses unique characteristics that render it useful in a variety of applications. Thanks to creative approaches to observing complex clinical data through the lens of mathematical coding, researchers have uncovered a crossroad between computer and medical sciences that offers an exciting landscape to improve the current clinical diagnostic and therapeutic approaches. Bringing together findings from multiple sources, such as private health information, laboratory, and physical examination, neural networks have yielded novel modeling systems in which different features in the medical dataset dynamically contribute to the maturation of the system's predicting and classifying functions. This potency is commonly attributed to the training function in a neural network, which enables the ANN to autonomously recognize the link between the input and outputs of a particular database. Besides the ANN's groundbreaking promises, a bulk of applications have surfaced its existing limitations, including local minima entrapment and extended processing time. In this context, evolutionary algorithms (EAs) are developed to address the ANN's shortcomings. In the present research, we recruit ANN-based BSA, VSA, and GOA algorithms to optimize the neural network's prediction competence. The proposed models are utilized in a database from UCI databank to predict the outcome of bone marrow transplant in children with hematologic conditions. Root-mean-square deviation (RMSD), ROC, and AUC measures are harnessed to analyze and compare the outcomes of different models. The ANN-BSA model is recognized to bring about the most viable results concerning the relationship between input and output layer data (that is, clinical features and survival rates). This research provides solid proof of the significant assistance of ML systems to healthcare practitioners to estimate an individual-based prognosis.