Artificial Neural Network Algorithm on e-Nose Devices for Honey Classification

Determining the type of honey is an essential step in maintaining the authenticity and quality of the product. This study developed an electronic nose system based on MQ-3 and MQ-135 gas sensors that record three main volatile parameters, namely carbon dioxide, acetone, and alcohol. A total of 541 data samples were normalised using the min–max method, then divided using a 75 per cent hold-out scheme for training and 25 per cent for testing. The classification model used a multilayer perceptron artificial neural network with a 3–7–3 architecture, Adam optimiser, learning rate of 0.001, batch size of 32, and 1000 epochs. Testing results on 135 test samples showed an overall accuracy of 88.89%. The evaluation per class shows that forest honey achieved 100% precision, 100% recall, and an F1-score of 100, cultivated honey achieved 97.1% precision, 70.8% recall, and an F1-score of 82.1, while trigona honey achieved 75.0% precision, 97.7% recall, and an F1-score of 84.8. These findings indicate that the combination of e-nose and JST is capable of identifying honey with a high level of accuracy, while also opening up opportunities for the application of this method as a rapid detection system to support the authenticity of honey products.