Combining classifier ensembles with fuzzy clustering to predict online food sales

Abstract

With the development of information technologies, data volumes have increased and the processing of numerous data has become a necessity for handling the competition, especially for food sector. In this context, hourly online order demand was estimated from 34 restaurants in fast food sector covering the period of January 2016–September 2017 by using as Logistic regression+Adaboost algorithm, SVM+Adaboost algorithm, NARX+Adaboost algorithm, ARIMA+Adaboost algorithm and Random Forest+Adaboost algorithm in crispcollective learning models. In addition to that case, fuzzy SVM, fuzzy ANN and fuzzy logistic regression and fuzzy random forests are also applied for the further analysis of fuzzification of classification process. According to the estimation parameters (Sensitivity, Specificity and Accuracy), fuzzy random forests technique was found to provide the best predictive performance. This better results can be thought from the aggregation rule that depend on Hamming distances which assists the closer predictions are given as the higher weights.