ABSTRACT

Algorithms for machine learning have been commonly used in many applications and fields. Diseases like malaria can be predicted using machine learning algorithms to create a suitable and precise model that predicts when, how, and where the outbreak will occur. The model performance can be improved using various strategies, such as ensemble methods or fine-tuning the hyper-parameters. Choosing a proper hyper-parameter configuration impacts the model accuracy, so it must fit well to improve the performance even though different optimization techniques exist, each with its benefits and disadvantages when used to solve various tasks. When working with hyper-parameter optimization strategies, a greater understanding of how machine learning models work is usually necessary. In this study, a hybridized form of Random-Grid optimization is proposed and applied to the three standard machine learning algorithms; K-nearest Neighbors, Random Forest, Support Vector Machine and their ensemble. It works by doing a randomized search to obtain the best hyper-parameters and then using it in a grid search to minimize the time it will take to search for the optimal combination of hyper-parameters. The grid and the random search work together to obtain the optimal hyper-parameters to improve the model accuracy by specifying the number of iterations to be performed when looking for the optimal model. The proposed technique is compared with well-known techniques such as Bayesian optimization, the Grid search, Random Search, genetic algorithm, and the particle swamp and evaluated using the malaria dataset obtained from the World Health Organization. The proposed technique improved the prediction accuracy for each base learner through an experimental study, and the ensemble method gives better results for HRGO-ensemble, HRGO-RFC, HRGO-SVM, and HRGO-KNN with 97%, 96%, 95 and 93%, respectively.

Keywords: Malaria, Machine Learning Algorithms, Hyper-parameter, Genetic Algorithm, Particle Swarm Optimization, Bayesian Optimization.