A novel fuzzy neural ensemble decision support system for whisper cancer analysis

Abstract

Early detection of ovarian cancer is critical to allow women to seek treatment when the cancer is still in the early stages, which will increase their chances of survival. Unfortunately, detection tools available now are either too expensive or they do not have enough sensitivity and specificity.

The difficulty of finding a good detection tool is further increased by the presence of unavoidable missing data. Currently there are no suitable method of handling missing data which can be employed easily while maintaining the accuracy and sample size. The derivation of a good missing data handling method will allow experiments to proceed smoothly with an increased precision and reliability of the final cancer detection result.

7 different classifiers were tested out in this project for their accuracy in classifying 145 samples of ovarian cancer. As ensemble learning has the capability of reducing the likelihood or a poor selection, results of different models within each classifier were combined for ensemble learning. Ensemble learning demonstrated its strength by improving accuracy by nearly 7%.