An Embedded Machine Learning-Based Spoiled Leftover Food Detection Device for Multiclass Classification

Abstract

Food waste’s negative environmental repercussions are causing it to become a global concern. Several studies have examined the
factors influencing food waste behaviour and management. This work was motivated by the lack of previous research on machine
learning and electronic noses to detect contamination from leftover cooked food. This work proposes using machine learning algorithms and electronic nose technology to recognise and forecast the contamination in leftover cooked food. After five days of storage, the freshness of cooked leftovers was evaluated using an electronic nose combined with machine learning algorithms. Most food samples used in this work were from Malaysian’s leftover lunch and dinner dishes. Four (4) gas sensors—MQ-2, MQ-136, MQ-137, and MQ-138—are used in developing the electronic nose to identify the presence of gas in the food sample. The data from the gas sensors was analysed using machine learning methods, namely Random Forest, k-nearest Neighbors, Support Vector Machine, and Linear Discriminant Analysis. Based on the results, a multi-classification technique yielded a greater accuracy rate in classifying and identifying the level of contamination in the cooked food leftovers, with average accuracy ranging from 90 percent to 100 percent. In conclusion, the work demonstrates a novel method for using machine learning algorithms to classify, identify, and predict the contamination level of leftover cooked food, contributing to reducing food waste generated primarily by Malaysians.