Future Sales Prediction Using Regression and Deep Learning Techniques

Sales prediction can help companies to manage resources more effectively, such as cash flow and production, and create better business plans. Sales forecasting allows the company to be proactive instead of reactive. It can help a company in adopting a suitable production policy. It also helps in controlling the inventory and cutting costs on machinery and labor power. In this paper, we examine and compare some machine learning models for predicting future sales, namely Light Gradient Boosting Machine (LGBM), Stacked Long Short-Term Memory (LSTM), and Multi-Layer Perceptron (MLP) with LSTM encoded features. The models were evaluated on the dataset containing historical sales data of Russian Software Firm -1C Company. The results manifest that LightGBM performs better than the other two models with 13.24% less Root Mean Square Error (RMSE) compared to stacked LSTM and 26.3% less than MLP with LSTM encoded features.