Adaptive Model to Estimate Most Significant Features for Oversampling Medical Data

Abstract

Computerized classification plays an important role in the classification of cancer stage. Hence, there is a growing need for automatic classification of cancer data. In this paper, a new model for stage classification of cancer data is developed. The proposed model use decision tree data mining technique that is trained on medical data to classify cancer data into several stages based on the weights of the most significant features. Hence, this paper proposes boost strapping approach for generating data oversampling to solve the problems of medical data scarcity. Ultimately, the proposed model improve the gain ratio technique to predict the weights of the factors (attributes) that affected in the staging of each patient case before and after oversampling. The performance of the proposed model is evaluated to develop more cost-effective and easy to use systems that support clinicians. The experimental results show that the proposed model precision is 94% for the original dataset and 90% for the oversampling dataset. The result illustrates the promising capabilities of the model for detecting breast cancer stages by minimum data set and minimum attributes.