A Seasonal Approach for Analysis of Temporal Trends in Retail Marketing using Association Rule Mining

Abstract

In an era when the customer is deemed as the king of the retail market, it pays heed to analyze every dimension of the customer's purchase behavior to provide an insight into their buying patterns and to better understand retail rationale of the customers. This dissertation has attempted to envisage the temporal traits of the customer's behavior in the Retail marketing. The research has put forth a seasonal study on the association amongst products in the realms of Retail Industry. In a country like India, obsessed with diverse seasons, the companies need to come up with better seasonal strategies that drive the market. Seasons have various dimensions to it viz. climatic seasons, festival seasons, etc. For each of those seasons the target audience keeps changing and the association of a set of products with customers also changes. This dissertation carries out an empirical study on the impact of seasonal and socio-economic factors on the buying patterns of the customers using Data mining tools and specifically applying the association rule for the products sold in the retail market sector. The results were analyzed by segregating the dataset into three seasons namely, January-April, MayAugust, and September-December. The results were interpreted with relevance to the seasonal behavior of the customers defined by the association of products. Weka 3.7.9 data mining tool is used for analyzing the data collected from XYZ supermarket located in Kanchipuram, TamilNadu, India. A mammoth 12000 transaction dataset and 215 product categories were involved in the research. This study reveals threadbare about the temporal association of products, for the Retail market to capitalize upon and to establish a better understanding and to enhance their customer relationship. General TermsAssociation Rule Mining in Retail Marketing.

OMBA: User-Guided Product Representations for Online Market Basket Analysis

“…Because of the low support of the latter, conventional MBA techniques could fail to check the association between these two products despite them having a strong association with a Lif t score of 43.45. Capturing associations of products that have lower Support values is important due to the power law distribution of products' sales [13], where a huge fraction of the products have a low sales volume (as depicted in Figure 1b).…”

“…Third, conventional MBA techniques consider the set of transactions as a single batch to generate association rules between products, despite that the data may come in a continuous stream. The empirical studies in [18,13] show that the association rules of products deviate over time due to various factors (e.g., seasonal variations and socio-economic factors). To further illustrate this point, Figure 1c shows the changes in sales of two product categories over time at a retailer.…”

“…Conventional MBA techniques fail to capture these temporal variations. As a solution to this problem in [18,13], the transactions are divided into different time-bins and conventional MBA techniques are used to generate association rules for different time bins from scratch. Such approaches are computationally and memory intensive; and ignore the dependencies between consecutive time bins.…”