ABSTRACT
Cancer is one of the most deadly diseases in the world over. In developing countries, the treatment and diagnosis of cancer is still a herculean task with regard to prediction. This is due to the late diagnosis of the disease, brain drain, paucity of practitioners, scarcity of medical equipment to use, and lack of political will by the government to deal with the myriad of problems associated with it. Over the years, specialists in both the medical and computer fields have been working on using computational intelligence - machine learning algorithms - for cancer prediction as a second opinion for decision-making. Although there have been studies showing improvement in their use of machine learning techniques, there is still no one-size-fits-all model that addresses all cases of diseases. The future of cancer prediction and treatment is a crucial area of research, especially in developing countries with limited medical infrastructure. The use of computational intelligence and machine learning algorithms has shown promise in improving cancer prediction as a second opinion for medical decision-making. This study applied six feature selection methods comprising information gain, gain ratio, reliefF, chi-square, FS-P, and mRMR. The classifiers used for the classification are SVM, naive Bayes, decision trees, K-NN, multilayer perceptron, and logistic regression. The ensemble methods are bagging, boosting(AdaBoost), and stacking methods. The work shows that FS-P and mRMR have improved performances in terms of the number of features selected, time complexity, and accuracy, with accuracy levels reaching between 90.13% and 96.25%. This project highlights the potential for using ensemble methods to address some of the challenges in cancer prediction and provides roadmaps for improvement in the future.
