AN APPLICATION OF NA¨IVE BAYESIAN CLASSIFIER FOR STYLES IN APPAREL INDUSTRY

Abstract

This study introduces a quality indicator for the apparel industry that can be used to identify the critical styles out of a new set of styles. This indicator can be incorporated in designing the production procedure by taking proactive decisions on the critical styles. The study presents a Bayesian classification in the context of supervised learning. In particular, na¨ıve Bayes algorithm is employed to classify the apparel styles when a new set of styles is received to the company. Pareto chart is then used to identify the critical styles based on 80% rule. In order to apply the procedure, the end-line quality inspection data for 80 styles of 4 different brands were collected for 4 months. The defect/non-defect status and corresponding defect categories for each style were recorded. Style descriptions for all styles were collected from customer specification sheets. In order to reflect the key qualities of the style, the variables for the style descriptions were chosen. Afterwards, the classification model was applied to the training set of 60 style descriptions, and the predictions were made for the rest of the styles. Based on the “confusion-matrix” of the classifier, obtained from Python programming, the final precisions of the classifier for brands Lululemon, Lacoste, CK, and Nike were 0.85, 0.50, 0.85 and 0.50 respectively, with a 0.875 overall accuracy of the classifier. For an apparel manufacturer, what the study proposes is of great importance, for it enables saving time, cost, labour, and most importantly the quality, while being efficient in production.