Prediction of Monthly Total Sales for a Company using Deep Learning

Abstract

The popularity of deep learning in predicting the monthly total sales for a company is the current research trend in artificial intelligence. Deep learning, Long Short-Term Memory (LSTM), with the ability to remember the past, has replaced the traditional, Auto-Regressive Integrated Moving Average (ARIMA) in prediction. However, whether conventional or deep learning has the best accuracy is still an unanswered question arising from advancements in computer computational power and modern machine learning algorithms. This project compares the accuracy to determine the best algorithm to forecast the company’s future three months' total sales. The monthly sales of companies are collected and pre-processed. ARIMA estimator and Keras API are used to construct the models, Root Mean Square Error (RMSE) for measuring accuracy, and successfully predicted the future three months total sales using Google Colaboratory in Python. LSTM outperforms ARIMA by obtaining a percentage error ranging from 8.84 - 12.64 %, whereas ARIMA's percentage error ranges from 71.60 - 85.84 %. Deep learning as the state-of-art in predicting monthly total sales aids the business leader in making wise business plans.