Publication: Machine learning algorithm for sales forecasting
Loading...
Date
2024-07
Authors
Lai, Kar Wai
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Sales forecasting holds significant importance in business management, exerting substantial influence over resource allocation, marketing strategies, and
financial decision-making, thereby being pivotal across various facets of business operations. Hence, developing a robust machine learning algorithm for sales
forecasting is imperative. This paper aims to develop a prediction model capable of forecasting sales in six months rather than daily, weekly, or quarterly by employing Python-programmed algorithms to attain high accuracy. The study involves the development of four distinct sales prediction models utilising machine learning techniques such as Random Forest regression, Light GBM regression, autoregressive integrated moving average (ARIMA) model, and long short-term memory (LSTM) model. The sales dataset used for training and evaluation of these models was obtained from Kaggle covering the period from April 2018 to December 2022. The accuracy of each prediction model is assessed using four performance metrics which are Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared value (R2 score). The best prediction model for predicting monthly retail sales is the Light GBM model with an 82.3% accuracy based on MAPE score. To ensure the chosen sales prediction model’s performance is consistent and reliable for future sales forecasting, the Light GBM model undergoes a validation process involving hyperparameter tuning and the selection of random points from the dataset to assess the model's performance. The monthly sales trend for the next 6 months forecasted by the Light GBM model provides a valuable resource for smartphone retail outlets to effectively plan their finances for the future.