Parallelizing Active Memory Ants with MapReduce for Clustering Financial Time Series Data
Ying Ying Liu, Parimala Thulasiraman, Ruppa K. Thulasiram · 2016
Clustering financial time series data is a computationally intensive problem. Ant brood clustering (ABC) is a meta-heuristic algorithm inspired by how ants sort the broods in their nest according to their shape and size. In this paper, ants are incorporated with short term memory to avoid redundant random walks, a drawback experienced in original ABC. This algorithm, called ABC-INTE (Ant Brood Clustering with Intelligent Ants), is applied to cluster financial time series data using MapReduce programming model. Our algorithm employs alternate number of mappers over multiple MapReduce iterations to exploit data parallelism and indirect communication among mappers. We evaluate the clustering quality, by using the sum of squares for the mean intra-cluster distance (MICD) to measure the intra-cluster distance and sum of squares of the inter-clusters distance (SSICD) to measure the inter-cluster distance. Our algorithm achieves a value of 475.47 for the ratio SSICD/SS(MICD), which outperforms both sequential implementation and the MapReduce implementation with multiple mappers but no multiple MapReduce iterations.