Elephant herding optimization technique based neural network for cancer prediction

Abstract

Cancer is a kind of uncontrolled growth of abnormal cells in any part of the body. It can be of many types. Early prognosis of cancer is the only way to treat it in a better way for researchers. It is very important to classify cancer in high or low risk group and that can be done by applying different machine learning techniques. In this study, a nature-based machine learning technique is developed named as Elephant Herding Optimization algorithm (EHO) which is validated using some cancer datasets like lung cancer, breast cancer, and cervical cancer. Here, the feature selection algorithm like ANOVA and Kruskal-Wallis tests are used where the relevant number of features are selected. The performance of EHO is evaluated using Root Mean Square Error (RMSE) and Correct Classification Rate (CCR) with and without feature selection. The RMSE value of the EHO algorithm is compared with Local Linear Wavelet Neural Network (LLWNN) and Particle Swarm Optimization (PSO). EHO shows 0.9837 CCR in the breast cancer dataset, 0.9671 CCR in the cervical cancer dataset, and 0.8821 CCR in the lung cancer dataset using ANOVA test i.e. the best result in comparison to other optimization algorithms. According to the tables it is clear that classification techniques takes more time without feature selection techniques but less time with feature selection techniques.

Keywords

• Local Linear Wavelet Neural Network (LLWNN)
• Elephant Herding Optimization (EHO)
• Artificial Neural Network (ANN)
• Particle Swarm Optimization (PSO)
• ANOVA