Application Of Machine Learning-Based Approach For Humidity Prediction In Food Drying: A Case Study From Nestlé Coffee Factory

Abstract

Coffee is one of Vietnam's important export products, especially for non-caffeinated coffee products, Nestlé Vietnam is proud to be the largest supplier in Asia. The final moisture content of coffee beans is an important factor affecting the quality of the product as well as the economic value of this item, in addition it also has an impact on environmental and social factors. With the remarkable development of physical-system communications, it helps to obtain a lot of real-world data, but traditional methods often ignore the value of this valuable source of information. The objectives of this thesis research is to apply machine learning models, specifically ANN and ANFIS models combined with GA algorithms, to take advantage of large data collected about the industrial environment to predict humidity of coffee beans during the drying process. Actual data from the Nestlé coffee factory is used to train the machine learning model, as well as test the effectiveness of the prediction. The model is then validated using validation metrics such as RMSE, MAE and actual model training time. The results obtained demonstrate the effectiveness of applying machine learning models in research, specifically the GA-ANFIS model shows more optimal performance than the traditional ANN model in both performance, training and solving time. The study also analyzed the impact of input features and identified list of features that need more attention, hyper care and control during the drying process.