Multiagent System for Pattern Searching in Billing Data

Abstract

In this paper we present an agent-based pattern searching system using a distributed Apriori algorithm to analyse billing data. In the paper, we briefly present the problem of pattern mining. Next, we discuss related research focusing on distributed versions of Apriori algorithm and agent-based data mining software. Paper continues with an explanation of architecture and algorithms used in the system. We propose an original distribution mechanism allowing to split data into smaller chunks and also orthogonally distribute candidate patterns support calculation (in the same computation task). Experimental results on both generated and real-world data show that for different conditions other distribution policies give better speedup. The system is implemented using Erlang and can be used in heterogeneous hardware environment. This, together with multi-agent architecture gives flexibility in the system configuration and extension. © Springer-Verlag Berlin Heidelberg 2013.