Companies aim to offer customized treatments, intelligent care, and a seamless experience to their customers. Interactions between a company and its customers largely depend on the company’s ability to learn, understand, and predict customer behaviors. Customer behavior prediction is a pivotal factor in improving a company’s quality of services and thus its growth. Different machine learning techniques have been applied to gather customer data to predict behavioral patterns. Traditional methods are unable to discover hidden patterns in ideal situations and need to be improved to produce more accurate predictions. This work proposes a novel hybrid model comprised of two modules: a novel clustering module on the basis of an optimized fuzzy deep belief network and a customer behavior prediction module on the basis of a deep recurrent neural network. Customers’ previous purchasing characteristics and portfolio details were analyzed by applying learning parameters. In this paper, the deep learning techniques were optimized by applying the butterfly optimization method, which minimizes the maximum error classification problem. The performance of the system was evaluated using experimental analysis. The proposed approach was compared to other single and hybrid-model-based approaches and attained the highest performance in the respective metrics.

Nowadays, most companies are utilizing customer behavior mining frameworks to improve their business strategies. These frameworks are used to predict different business patterns, such as sales, forecasting, or marketing. Different data mining and machine learning concepts have been applied to predict customer behaviors. However, traditional approaches consume more time and fail to predict exact user behaviors. In this paper, intelligent techniques, such as fuzzy clustering and deep learning approaches, are utilized to investigate customer portfolios to detect customers’ purchasing patterns. To accomplish this objective, hierarchical fuzzy clustering was applied to compute the relationship between products and purchasing criteria. According to the analysis, similar data are grouped together, which reduces the maximum error classification problem. Then, an optimized deep recurrent neural network is incorporated into this process to improve the prediction rate. The discussed system efficiency is evaluated using a number of datasets with respective performance metrics. The proposed approach was compared to other single model-based and hybrid model-based approaches and was found to attain maximum accuracy and minimum error rate in comparison.