Using Machine Learning Algorithms for Cancer Image Dataset: A Predictive and Prescriptive Analysis

This paper focuses on the problem of using machine learning techniques on cancer/tumor prediction. One of the biggest applications of big data and machine learning is in the field of medical domain. Consequently, a health care organization that uses the techniques of machine learning and big data to treat patients see fewer mishaps or gets enough time to deal with them in advance. Classification and prediction of the images are fairly easy task for humans, but it takes more effort for a machine to do the same. Machine learning helps to attain this goal. It automates the task of classifying a large collection of images into different classes by labelling the incoming data and recognizes patterns in it, which is subsequently translated into valuable insights. Furthermore, the prediction of Osteosarcoma case for one of the four classes of tumor namely Non tumor, Non-Viable tumor, viable tumor, Viable: Non-Viable tumor has to be done. The quantitative analysis is done using various machine learning libraries of python. The three classification algorithms used for image analysis are random forest, SVM, and logistic regression. The metrics used for performing perspective analysis are precision, recall and F1 Score. The domain of medical imaging helps providing important information on anatomy and organ function subsequently detecting disease states. Although the characteristics of medical data make its analysis a big challenge notwithstanding that machine learning techniques could make the task easier. The results show that the random forest algorithm has performed best amongst the three classification algorithms when given with less complicated scenario, with prediction accuracy, precision, recall and f1 score of 100%. But the performance of every classification algorithm degrades when provided with the cases of Osteosarcoma which has got more complicated scatter graph. However, the logistic regression retains its performance by predicting tumor cases with 99% accuracy. For future scope, various other machine learning algorithms can be applied to observe their performance on the same set of features extracted from the cancer image data set.