Demand Forecasting Of Inventory For R Company Based On Machine Learning Theory

It is an extremely important aspect for enterprises to occupy a place in the fierce market competition and study how to improve the logistics management level.In the logistics management activities of enterprises,inventory management is very important.Due to the two opposite objectives of inventory management,"reducing cost" and"avoiding shortages",forecasting inventory demand accurately is the basis and premise of inventory management.It directly affects the profit of the enterprise and the service level of the customer.In the context of the blowout growth of large global data,enterprises have more and more information data.The prediction of inventory demand is no longer a single fitting of time series based on historical sales data.We need to build a more scientific and comprehensive prediction model based on the prediction of complex data including historical sales data,so as to achieve more efficient and accurate prediction goals.Machine learning model has better performance in dealing with complex and huge data than traditional prediction model.Therefore,more advanced machine learning theory is introduced into inventory demand prediction,which can achieve better prediction results.Based on 149,740 historical data samples from 200 stores of R company in the past 31 months,three inventory demand forecasting models were established based on stochastic forest and generalized regression neural network respectively.The results show that the random forest model has good noise data recognition ability,high prediction accuracy,strong generalization ability,low complexity of parameter adjustment and fast training speed.The predictive effect of the generalized regression neural network model is worse than that of the random forest model.Although the ability to recognize the "noise" data is poor,the generalization ability of the model is very strong,and it is very good in the random selected test samples.The training speed of the model is fast and the parameter adjustment is difficult.However,it is very high requirement for the computer hardware to deal with the large sample.The prediction accuracy of the support vector regression model is poor,the curve of the prediction result is too slow,the generalization ability is poor,and the time needed for the training is too long.It is less efficient than other models.At the same time,through the prediction results from random forests,we can conclude that the most influential feature of R shops is shop numbers and promotions,and competitors and different weeks also have a greater impact on demand data.Then,Frank and Wolfe algorithms are used to determine the weight of combined prediction.The results of combined prediction are compared with the prediction results of a single model.The results showed that combination forecasting can achieve a better prediction accuracy than a single model.The study of this paper proves that machine learning algorithm can achieve good results in the prediction of sample set,proving that the machine learning algorithm has a good application prospect in dealing with the problem of inventory demand prediction based on large sample data.