An Association Based Approach to Elicit and Measure Impact of Features on Sales of a Garment Retail

Abstract

In this article, an association-based approach is proposed for determining the feature importance of a given dataset which includes the target variable. In particular, the concept of Market Basket Analysis (MBA) is applied for enumerating the relationships between the target variable and each of the features which lead to the importance of those. Mention that the MBA is generally used for obtaining the recommended items based on the togetherness of the items. Nevertheless, an attempt is made in this paper to correlate the features given a target output by abstracting each feature to be paired with the target variable. The apriori algorithm and association rules are used for accounting for the coupling of features with the target feature. Precisely, Lift metric of MBA is the key to computing the associativity in this context. That is, each feature’s importance is the sum of the individual ratio of Lift count of its values (observations) when paired with the target feature. The proposed methodology is tested on a dataset of a garment retail store that has information on several dresses. Each dress contains fifteen features including sales which is the lone numerical feature amidst the categorical features. Note that the sales are influenced by some of the features which generally the customers look for to prefer a particular dress over others. The results of the proposed methodology suggest that a couple of features are instigating sales at a higher rate than others. The outcome of the developed methodology is able to define a clear grouping of features according to the importance related to the target variable. The proposed methodology is applicable to a dataset where the feature selection is with respect to a target feature which is generally done in the case of supervised learning.