A team of scientists from the University of Johannesburg (UJ), collaborating across theoretical and experimental physics and computer science, have developed and trained a new machine learning (ML) technique to predict and diagnose diseases such as lung cancer, tuberculosis, cardiovascular diseases and malaria.

The results were published in IEEE Xplore in 2019.

The accurate diagnosis of diseases is critical to the survival of humans. At times, due to the overlapping nature and similarities of the symptoms, it can be challenging for an inexperienced clinician to properly diagnose diseases. A wrong or untimely diagnosis of disease can cost patients money, time and even death.

The UJ scientists have made a breakthrough in both technique and understanding. Using a suite of artificial neural networks (ANN) designed and trained to acquire knowledge about the task at hand, they have developed a new forest-building method for machine learning.

Lead author Professor Qing-Guo Wang, from the Institute of Intelligent Systems at UJ, says: “The case study on the diagnosis of autistic spectrum disorder shows that the proposed method achieves the prediction accuracy of the ensemble at above 96% with reduced variance, which is much better than those reported in the literature.

“In this new collaboration with Professor Tshilidzi Marwala, UJ’s vice-chancellor and principal, and Adeola Ogunleye, machine learning engineer, we combined decision trees and regression methods which are usually in two different branches of machine learning to take advantage of each.”

A number of intelligent systems integrate two or more AI techniques (ANN, SVM, KNN) with a fuzzy logic system to form a hybrid expert system (HES) that can take advantage of the various techniques.
The breakthrough in this research was in introducing randomisation at tree growth and forest creation.

The local prediction accuracies on the leaves are used to select a subset of the test data for actual predictions. The ensemble combines trees and gives a better performance than the individually best-performing tree.

By fusing tree-based machine learning with a random order the scientists believe that the symptoms of an ailment from a patient serves as the input vector to diagnose the ailment with the AI model.

“There is an urgent need for the development of easily implemented, automatic and effective screening methods. This will help health professionals and inform individuals whether or not they should pursue a formal clinical diagnosis,” according to a statement from the UJ team.