Optimizing Sales Forecasting in Dairy Industry Using Neural Networks

This study attempts to design a more accurate and reliable sales forecasting model for the dairy business, specifically to mitigate challenges in Al-Youm Food Company's case.The nature of the dairy business, volatile demand, and the perishable nature of their products demand accurate forecasting to help restrict waste and maximize profits.The company is employing classic forecasting methods, which failed to capture their sales patterns' complexity.In response to this, a neural networkbased forecasting model was developed and trained through Python machine learning packages using 43 weeks of sales history (January -October 2023).The neural network was compared to classic forecasting methods, i.e., exponential smoothing and moving averages.The neural network model was found to be more accurate in prediction, having a Root Mean Squared Error of 19.68 compared to 20.52 for exponential smoothing and 21.98 for moving average.Similarly, the Mean Absolute Percentage Error for the neural network was found to be 27.76%,better than that of exponential smoothing (29.33%) and moving average (30.06%).The stability of the model was also verified using training loss and validation loss analysis.An extended forecast was developed, forecasting sales between week 44 to week 80.With limitations such as short periods of historical data and lack of extraneous variables, it is found that neural networks can be employed to improve sales forecasting in the dairy business significantly.The application of this model in Al-Youm Food Company is likely to lead to better control of inventory, minimized waste, and higher profits, and even be used as a model for other companies in the business to be more efficient and more sustainable in their business in the dairy business.