Spark-Based Parallel Method for Prediction of Events

Prediction of events is imperative in many areas of social network (SN) applications. These events influence the temporal evolutionary characteristic of social networks. A study of these events can give better insights to understand the evolutionary patterns (communities) in social networks. One of the major challenges in such implementation is the processing and structuring of large datasets to suit ML models. This paper proposes a Spark-based parallel method for detection, mining, and prediction of these events that influence the evolution of communities in a temporal SN. The proposed framework processes large temporal data (taken from the DBLP dataset), uses parallel algorithms to detect the structural changes, and applies ML techniques to predict the future structural changes (events). The proposed methodology uses ensemble ML methods in the Spark ML pipeline to achieve the desired performance and accuracy. The experimental results justify that the proposed framework can predict future events with an accuracy of 82% and saves 99% of computational time.