Regularized Logistic Regression Model for Cancer Classification

Abstract

Cancer is a serious disease and is considered one of the causes of death. Making it worse, many cancers are diagnosed too late. Early, diagnosis of cancer helps in taking correct steps towards treatment. This paper introduces a machine learning model to diagnose and classify different types of cancer. This model is implemented based on regularized logistic regression. The regularization techniques L1, L2 and Elastic Net are evaluated where L2 outperformed other techniques. Also, the proposed model is optimized using Stochastic Gradient Descent (SGD) and Averaged Stochastic Gradient Descent (ASGD) where ASGD outperformed SGD. The results showed that the model with best performance is obtained with L2 regularization when optimized with ASGD. The best model performance is evaluated using cross validation yielding 99.6%, 90.27% and 98.08% test accuracy for Ovarian, Colon and WBCD data sets respectively.