Real-Time Sales Forecasting Using Prophet and Gradient Boosting Models

Abstract

The accuracy of sales forecasting plays an essential role in optimizing operations and reducing inventory costs while increasing profits. However, the application of traditional methods such as ARIMA might be ineffective for predicting complex trends, seasonality, and real-time data. The research paper offers a new solution to the real-time sales forecasting problem by proposing a hybrid model that combines the Prophet time series method and the gradient boosting algorithm. The Prophet model will help address issues related to trend and seasonality, which will be supported by the use of gradient boosting algorithms, including XGBoost, for more accurate predictions based on advanced feature engineering and ensemble learning. In this study, the hybrid model is suggested to make the most of both approaches and better account for the dynamics of sales data. The paper uses historical sales data obtained from a retail store and takes into account all possible variables that might influence sales data. These factors include both internal and external variables. The performance of models is evaluated based on commonly used measures such as MAE, RMSE, and MAPE. The findings from experiments prove that the hybrid model is superior to both Prophet and the gradient boosting model, with an MAE of 4.5%, an RMSE of 6.2%, and an MAPE of 5.3%. These improvements are significant compared to the predictions from the ARIMA model. It is also evident that the proposed technique shows a promising real-time prediction performance by making decisions based on real-time data instantly. The findings show that the Prophet model is highly effective at predicting sales in real time, particularly in highly volatile and fast-changing markets. Further research will focus on scaling up the proposed model and applying it in other industries facing similar forecasting challenges.