STARM: STreaming Association Rules Mining in High-Dimensional Data

Predictive analytics involves using Data Mining algorithms to discover knowledge from large databases. The Association Rules (ARs) mining technique is considered to be one of the most prevalent data mining techniques in this context. When it comes to Big Data, we talk about data stream mining which is the process of extracting knowledge from continuous data streams. In this paper, STARM (STreaming Association Rules Mining) is proposed as an efficient and distributed algorithm for mining ARs. Based on the transaction-sensitive sliding-window model, the Apriori algorithm is applied to data streams to extract frequent itemsets (FI) that are then generated into ARs via Spark streaming framework. A Dimensionality Reduction (DR) step takes place as a data preprocessing step that may reduce the search space. The conducted experiments show that the proposed streaming model achieves state-of-the-art performance.