Application of Meta-Optimization to Improve Big Mart Dataset Predictions

Abstract

Multiple studies have been conducted to get the best classifier machine learning model, using various open source datasets. The dataset used in this study is Bigmart data set with 8524 samples. There are 17 variables in this dataset that can be used for the study. There are a lot of available studies performed on variables like health status variable, item type and store size variables. These studies have performed variations of machine learning models to get the best possible prediction. This study is a follow up study of the one done previously, where nine classifiers were applied on bigmart dataset and OneR classifier performed the best for predicting item type variable giving 84.46% accuracy. In this study, to improve the performance of the classification, four meta-heuristic optimizers, namely Elephant Herding Optimization, Monarch Butterfly Optimization, Harris Hawks Optimization and Slime Mould Algorithm are employed that assist hyper parameter tuning in a systematic way. It was concluded that Elephant herding optimization improved the OneR classification to 95.2181%. This increased accuracy is a significant improvement from the previous values.