This research developed an automatic classification system to determine the freshness level of beef using the MQ-135 gas sensor, which detects the concentration of ammonia and carbon dioxide as the primary indicators of spoilage. Additional sensors, MQ-3 and MQ-2, were used to detect ethanol and butane, ensuring that no external gas contamination occurred during the data collection process. Data were collected over thirty-five days and manually classified into three categories of freshness: fresh, less fresh, and not fresh.

Due to an imbalance in the number of data points across categories, the RandomOverSampler method was applied to balance the class distribution. The Support Vector Machine algorithm was used as the classification model, and optimisation was carried out using GridSearchCV. The final model demonstrated an accuracy of 95.24%, with excellent performance in distinguishing between fresh and non-fresh classes, and improved recognition for the less fresh category.

Six experts conducted validation through assessments of the colour, odour, and texture of the meat. The validation results indicated a high level of consistency between the system output and expert assessments. In addition, visualisation of gas concentration trends during the storage period reinforced the relationship between increasing gas levels and decreasing freshness. The integration of multi-gas sensors with machine learning algorithms resulted in an objective, non-invasive, and real-time monitoring system. This approach can be implemented in the food industry distribution chain to enhance safety, reduce spoilage, and support sustainable quality control.