Optimizing Predictive Modelling of Customer Behaviour Using Simulated Annealing

Abstract

Customer behaviour modelling is an important data mining approach for making operational and strategic decisions. Market Basket Analysis (MBA) is used to determine products that customers purchase together. Knowing the products that a customer will shop as a group is very helpful to a retailer. The business can use this information in predicting sales at the right time, at the right place, and for the right customer. Moreover, Company Marketers can use the basket analysis results to determine new products to serve their existing loyal customers. In the present paper, Genetic Algorithm is used to determine the population of solutions, which consumes more time to produce a solution. In this paper, it is proposed to use the Extended HCleaner Algorithm to remove noisy data from the datasets. The pre-processed dataset is then submitted to an ANN model. From the ANN model, weights of the products are determined by an association. The products which get maximum weights are sent to Apriori algorithm, which calculates the optimal combination of the products. The solution of Apriori algorithm is sent to next level of the predictive model. It contains two stages: (1) Similarity calculation through cosine similarity technique and (2) Simulated annealing. The cosine similarity is used to obtain the association between products which is further sent to simulated annealing algorithm to find the single solution through association rules. Simulated annealing is used to minimize the response time. It is more effective than the existing system.