Machine Learning Algorithms for Medical Prediction: A Comparative Study

Abstract

The aim of the current study was to systematically compare the well-known machine learning algorithms in medical prediction for each clinical field. The scan compared supervised machine learning approaches (logistic regression, decision trees, random forests, support vector machines and gradient boosting machine) and deep learning including convolutional and recurrent neural networks.

The results indicated that chronic disease such as heart disease and diabetes are prone to be predicted well by the random forest and GBDT algorithms, whilst DNNs fare well for image and biosignal analysis. The research also alerted to the necessity of choosing an algorithm based on properties of the data and complexity and on the need for being able to understand the results, especially for medical purposes where explainability is a requirement.

This integration of deep learning algorithms with dimensionality reduction and unsupervised learning approaches may result in better and more interpretable predictive models and enhanced our explanation capability of these algorithms, particularly those to be used in clinic.