Data-Driven Sales Forecasting for Fashion Sales using Machine Learning Techniques

Abstract

Sales data classification is crucial for assessing business performance, predicting trends, and informing decisions. This study investigates the use of machine learning methods for sales data classification. Specifically, the focus is on utilizing supervised learning algorithms, such as Random Forest, Decision Tress, Support Vector Machines (SVM), and Long-Short Term Memory (LSTM) networks to classify sales data into different categories or predict sales patterns. The research begins with data preprocessing, including data cleaning, feature selection, and normalization. Amazon fashion data, along with relevant features such as main category, ratings, and number of ratings, discount price, actual price, and the sub-category of the fashion product are organized into a structured format suitable for machine learning algorithms. A comparative analysis is conducted to pinpoint strengths and weaknesses in sales data classification. The performance of every algorithm is measured using metrics, LSTM outperformed Random Forest and SVM with a maximum accuracy of 98%, sensitivity of 97%, specificity of 96%, and F1 score of 91%.