An improved weighted decision tree approach for breast cancer prediction

Abstract

The death rate in case of breast cancer existence can be reduced by identifying the tumour at an early stage. The survival rate can be increased if the tumour is identified initially and not spread to other organs. The mammography is able to recognize the various breast tissues with area size and criticality parameters. The machine learning algorithm can be applied on these breast tissue features to identify the chances of tumour recurrence. In this paper, a selective feature based improved decision tree algorithm is suggested to predict the chances of breast cancer occurence. Initially, each cancer descriptive symptom and features are processed under Chi square test to recognize the most contributing features. The ranked selected features are processed in the same order to generate the feature adaptive improved decision tree. For each tree node, the entropy and cost based rules are defined to predict the existence or non-existence of breast cancer. The proposed feature rank based improved decision tree is applied on two most popular breast cancer datasets taken from the UCI repository. The comparative results against the decision tree, naive bayes, random tree and random forest classifiers shows that the proposed model has predicted the breast cancer more accurately.