Evaluation of cigarette market state based on multi-source data modelling

Traditional cigarette market forecasting model usually has a low accuracy since it did not take the external data into account. Thus, a random forest was firstly used to extract features of data and rank the importance of influencing factors. Then, different external factors were eliminated, the percentage of reduced model interpretation was demonstrated, and expert feedback was introduced to input evaluation values. After optimising the training RF-LSTM model, the prediction of the whole market sales status were finally constructed, and the historical week cigarette market status evaluation model was also established. The proposed machine learning model had a high prediction accuracy and generalisation based on the local market data in province Guangxi of China. Overall results demonstrated that it can accurately and conveniently evaluate the market status of cigarettes.