Prediction time of breast cancer tumor recurrence using Machine Learning

Abstract

An in-depth study using the database from GLOBOCAN, CDC, and WHO health repository highlights the lethality of breast cancer, taking thousands of lives each year. However, a timely prediction of cancer can help patients to consult the doctor on time. In the past, various studies have successfully predicted the nature of the tumor to be benign or malignant and if the breast cancer tumor will reoccur or not but, no time-based models have been studied. With the help of Machine Learning, this study shows various prediction models that can be used to predict tumor reoccurrence time as accurately as 1 year. Among the 198 patients analyzed, 40% of the total patients were predicted to have breast cancer tumors reoccurring within 1st year of the diagnosis. The proposed machine learning techniques use various classification models such as Spectral clustering, DBSCAN, and k-means along with prediction models like Support Vector Machines (SVM), Decision trees, and Random Forest. The results demonstrate the ability of the model to predict the time taken by the tumor to reoccur or the time taken by the patient for full recovery with the best accuracy of 78.7% using SVM. This population-based study performed on multivariate real attributed characteristics data can therefore provide the patients a reasonable estimate about their recovery time or the time before which they should consult the doctor.