Prediction of Daily Ice Crystal Demand Using Modified Bi-xLSTM

Abstract

In this study, in addition to developing a predictive model based on Extended Long Short-Term Memory (xLSTM), an in-depth analysis was also conducted on the factors that can influence daily demand for crystal ice. Weather factors such as temperature, rainfall, and humidity, as well as Google search trends for “crystal ice,” were used as additional variables that are expected to improve the model’s prediction accuracy. The historical data collected consists of crystal ice sales over the past several months, including relevant external variables. Furthermore, this data was processed and cleaned using data preprocessing techniques such as normalization and handling of missing values to ensure it was ready for the model training stage.The implementation of the model involved splitting the data into training and validation sets, which provided an overview of how well the model is able to generalize to previously unseen data. The training process was carried out iteratively by tuning hyperparameters, such as the number of LSTM units, learning rate, and number of epochs, to achieve optimal model performance. Additionally, the binary classification feature integrated into this model allows SMEs to quickly decide whether additional production capacity for crystal ice is needed for the following day, based on the latest demand trends.Model evaluation was conducted not only based on accuracy values but also using a confusion matrix to gain deeper insight into the recall and precision rates. In this study, an accuracy of up to 90% was achieved, indicating that this model can be applied by various SMEs producing crystal ice to help increase their profits.