Quality assessment of chicken using machine learning and electronic nose

Abstract

Meat is highly perishable food and prone to microbial contamination under various storage conditions. Quality assessment at both retail and industrial levels often relies on organoleptic properties, gas chromatography, and total bacterial count, all of which require trained personnel and significant resources. As a result, there is a need for a more efficient and reliable system to determine chicken quality. This study investigates the use of an electronic nose system—a sensor array that detects odors and generates data, which is then analyzed by machine learning algorithms to predict chicken freshness. An electronic nose system was developed using six MQ gas sensors and one humidity temperature sensor. Data was collected from chicken samples over a period of 15 days. To evaluate the performance of the machine learning algorithms, different data splitting approaches were tested to understand their impact on model accuracy. Random Forest achieved 100 % accuracy with randomly split data and 69 % accuracy with non-randomly split data. Support Vector Machine, using the recursive feature elimination technique, attained 78.5 % accuracy without random splitting. The study also reviewed existing literature, highlighting that random data splitting is not suitable for electronic nose data. Overall, the findings suggest that the electronic nose system, combined with appropriate data handling and machine learning techniques, can effectively assess chicken freshness, potentially offering a valuable tool for the poultry industry.