BAT-ELM: A bio inspired model for prediction of breast cancer data

Medical informatics mainly deals with finding solutions for the issues related to the diagnosis and prognosis of various deadly diseases using machine learning and data mining approaches. One such disease is breast cancer, killing millions of people, especially women. In this paper we propose a bio inspired model called BATELM which is a combination of Bat algorithm (BAT) and Extreme Learning Machines (ELM) which is first of its kind in the study of non image breast cancer data analysis. The concept of BAT and ELM which has many advantages when compared to the existing algorithms of their genre have motivated us to build a model that can predict the medical data with high accuracy and minimal error. Here we make use of BAT to optimize the parameters of ELM so that the prediction task is carried out efficiently. The main aim of ELM is to predict the data with minimum error. For attaining a minimal error we have tested Wisconsin Breast Cancer Prognostic (WBCP) dataset upon three different learning functions (sigmoid, sin and tanh) and the function which produces the best result has been considered as the final. We carried out two case studies to support our model. In case study I the objective was to predict whether the breast cancer is recurrent or non-recurrent. The accuracy obtained for this case is found to be 95.7% with an RMSE of 0.32. In case study II our objective was to predict the time of recurrence, the result obtained for this case were found to be 93.75% accurate with an RMSE of 0.30. In both the cases tanh function performed better.