Scholarlink

Sales forecasting has become strategically very specific to the companies operational efficiency and a customer’s good experience in today’s retail analytics evolvement. In this work, I am going to dig into the effectiveness of a hybrid model that made up by the long short term memories(LSTM) holding pattern in time and the idea of XGBoost which is feature handling. The proposal entails an initiative to improve the precision and effectiveness of store sales forecasting. The aim of this study is to compare the predictability of the XGBoost, LSTM, and a mixture model using a multitude of dataset containing sales records data with a variety of characteristics. A modelling is designed where multiple performances metrics such as Root Mean Squared Logarithmic Error (RMSLE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R \(^{2}\) ) are used for performance comparison. The hybrid model however is a mixture between LSTM, which can handle sequential data very well and XGBoost that is known as a very efficient and quick feature analysis tool and is able to outdo the current sales forecasting as it stands. The research also shows that ensemble is an approach that tackles the weakness of single model but it forms the basis for the development of retail analytics using hybrid machine learning techniques.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid Approach Using LSTM and Gradient Boosting for Accurate and Robust Store Sales Forecasting

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shalim Sadman,
  • Doina Logofătu

摘要

Sales forecasting has become strategically very specific to the companies operational efficiency and a customer’s good experience in today’s retail analytics evolvement. In this work, I am going to dig into the effectiveness of a hybrid model that made up by the long short term memories(LSTM) holding pattern in time and the idea of XGBoost which is feature handling. The proposal entails an initiative to improve the precision and effectiveness of store sales forecasting. The aim of this study is to compare the predictability of the XGBoost, LSTM, and a mixture model using a multitude of dataset containing sales records data with a variety of characteristics. A modelling is designed where multiple performances metrics such as Root Mean Squared Logarithmic Error (RMSLE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R \(^{2}\) ) are used for performance comparison. The hybrid model however is a mixture between LSTM, which can handle sequential data very well and XGBoost that is known as a very efficient and quick feature analysis tool and is able to outdo the current sales forecasting as it stands. The research also shows that ensemble is an approach that tackles the weakness of single model but it forms the basis for the development of retail analytics using hybrid machine learning techniques.