Deep Learning Classification of Heart Disease Risk Using Evolutionary Hyperparameter Optimization

Ryan Kaddis, Devson Butani, Siri Sri Churakanti, Batyr Kenzheakhmetov*, Bhavesh Krishnaram Bhavesh, Prerak Patel, CJ Chung

Computer Science

Lawrence Technological University

ML, DL, HPO, Optimization

submitted by cj424

Hyperparameters are an often overlooked aspect of training neural networks. There is, however, an advantage to finding a set of hyperparameters that produce optimal model results. Using a dataset that classifies the risk of heart disease based on a number of physical parameters, optimal hyperparameters are found for a classification model using various evolution-based optimization strategies. These methods include Evolutionary Strategy (1+1) with 1⁄5 rule, Evolutionary Strategy (N+M) with 1⁄5 rule, and Genetic Algorithm (GA) using the DEAP library. Models produced using these methods consistently achieved accuracies above 80%, with one model optimized using a GA-based approach reaching 96.6% accuracy.

results from F2024 class

CC BY

Download File

Back to Search