Sales forecasting method based on LSTM and conformal forecasting

Sales forecasting is essential for market decision-makers. Accurate and reliable sales forecasts allow decision-makers to formulate effective market strategies in advance, yet few studies emphasize the reliability of these forecasts. This study proposes a method that integrates long short-term memory (LSTM) and conformal prediction to enhance the reliability of sales forecasts. First, the LSTM model is constructed, and data is fed into it. The model is then trained on the data, generating the predicted values. Next, the predicted values are fed into the conformal prediction model, which generates a confidence interval to represent their reliability. Finally, an experiment is conducted using a publicly available benchmark dataset from an e-commerce company. The method's effectiveness is evaluated by examining the neural network design, setting the loss function to quantile loss, and selecting the average coverage interval (ACI) predictor. The LSTM-conformal forecasting combination improves reliability, aiding decision-makers in developing more informed market strategies.