On the Scalability of Machine-Learning Algorithms for Breast Cancer Prediction in Big Data Context

Recent advances in information technology have induced an explosive growth of data, creating a new era of big data. Unfortunately, traditional machine-learning algorithms cannot cope with the new characteristics of big data. In this paper, we address the problem of breast cancer prediction in the big data context. We considered two varieties of data, namely, gene expression (GE) and DNA methylation (DM). The objective of this paper is to scale up the machine-learning algorithms that are used for classification by applying each dataset separately and jointly. For this purpose, we chose Apache Spark as a platform. In this paper, we selected three different classification algorithms, namely, support vector machine (SVM), decision tree, and random forest, to create nine models that help in predicting breast cancer. We conducted a comprehensive comparative study using three scenarios with the GE, DM, and GE and DM combined, in order to show which of the three types of data would produce the best result in terms of accuracy and error rate. Moreover, we performed an experimental comparison between two platforms (Spark and Weka) in order to show their behavior when dealing with large sets of data. The experimental results showed that the scaled SVM classifier in the Spark environment outperforms the other classifiers, as it achieved the highest accuracy and the lowest error rate with the GE dataset.