(Estimating Out-Of-Home Food Consumption Classes Using Artificial Neural Networks)

Abstract

Purpose - The purpose of the research carried out is to estimate out-of-home food consumption classes of Turkish households. Design/methodology/approach – The data used in this study were derived from the Household Budget Survey gathered by the TURKSTAT. The data obtained from a total of 11,521 households in 2019 were used in the analysis. Artificial Neural Networks (ANN) technique were used to determine out-of-home food consumption classes. Results - In the ANN method, 70% of the households were used to be trained for the model, and the remaining 30% were allocated for the test phase. The evaluation of binary classification problems and the performance of the model established were evaluated with the confusion matrix and the metrics obtained from this matrix. The model has an accuracy rate of 73.12% in the training phase and 73.39% in the testing phase. The sensitivity of the model is 77.28% in the training phase and 80.39% in the testing phase. The precision of the model is 71.22% in the testing phase and 69.73% in the testing phase. The metrics and comments on the metrics are explained in detail in the findings section. When the metrics calculated from the training and testing phase results of the established model were examined, it was seen that the system obtained correct and consistent results. Discussion - Apart from the variables obtained from the studies examined in the literature, it has been seen that the variables added to the model increase the classification accuracy of the established Artificial Neural Networks model. Almost all of the studies conducted in household out-of-food consumption in Turkey was carried out using econometric models. It is thought that the econometric models to be established with the new variables added will change the results of the previously obtained models. There’s no study that could be found in the literature that carried out to estimate household out-of-home food consumption classes using ANN method in Turkey. For this reason, this study makes an original contribution to the literature.