Prediction of Malignancy of Brain Cancer on SEER Dataset using Random Forest, SVM, and Naive Bayes Classifiers

Brain cancer is rarely occurring type of cancer. Survival rate of a person suffering from brain cancer is strongly dependent on type of tumour i.e. malignant or benign. Survival of patient rapidly decreases if tumour is malignant and has spread beyond primary location. Prediction of nature of tumour can help physicians to decide between targeted treatments for patient. Machine learning models are being widely utilized to improve the prognosis of cancer. This paper presents three different machine learning classification models to predict malignant or benign nature of brain tumour. Support Vector Machine (SVM), Random forest, and Naive Bayes classifiers are implemented using Python. Models are trained and validated on dataset obtained from Surveillance Epidemiology and End Results (SEER) for years 1973-2015. Performance of each classifier is evaluated and compared to other two classification models in terms of precision, recall and F-score. A tool is designed using classifier providing best performance on training dataset to predict malignant/benign nature of tumour for a test dataset provided as input to tool.