Association rule mining using hybrid GA-PSO for multi-objective optimisation
Aashna Agarwal, Nirali R. Nanavati · 2016
Association Rule Mining (ARM), a Data Mining process, extracts hidden strong relationships among a large set of the correlated data. With the burgeoning advancement and application of Association Rule Mining in diverse fields ranging from the web usage mining to medical diagnosis and business intelligence to geographical information systems, the decision-making in ARM involves a multi-objective perspective to obtain an interesting and accurate rule set. By considering the Pareto optimality, an optimal trade-off is established between the conflicting and incommensurate performance parameters - comprehensibility, interestingness and confidence of the mined rules. Both, Genetic Algorithm (GA) and Particle Swarm Optimisation (PSO), being population-based stochastic search method, have found their strong base in mining association rules. We propose an association rule mining scheme using our proposed multi-objective hybridisation of GA-PSO algorithm. The primary advantage of the proposed algorithm is that the hybridisation of multiple objective-GA with multi objective-PSO balances the exploration and exploitation tasks, resulting in valuable extraction of accurate and interpretable mined rules. Evaluating this hybrid model on Bakery dataset shows that with generation of comprehensible, interesting and reliable association rules, the model also converges four times faster than mono-objective hybridisation.