Classification of cooking oil aroma with high performance using chemometric and E-nose methods

Chemometric analysis using Linear Discriminant Analysis (LDA) and Artificial Neural Networks (ANN) was applied to differentiate the aroma profiles of cooking oil samples using an electronic nose. Five types of oils were tested, namely pure lard, pure olive oil, pure palm oil, lard-olive oil mixture, and lard-palm oil mixture. Each oil type was independently tested 40 times, resulting in a total of 200 measurements. The objective was to develop a method to detect lard adulteration in commercial oils. The LDA analysis revealed that the first discriminant function accounted for 59.6% and the second for 34.5% of the total variance. Olive oil was classified with 100% accuracy, palm oil with 90%, while the remaining 10% overlapped with the lard-olive oil mixture (50 : 50 v/v ratio) group. Pure lard and its mixtures were relatively more difficult to classify accurately. Using the backpropagation ANN method, 94% of the test samples were correctly identified. The ANN model employed 50 training data, 100 testing data, and several data points for cross-validation. The results demonstrated stable weight convergence with outputs closely matching the target values, indicating a low level of overfitting. Optimal performance was achieved at the 185th epoch, with a learning rate of 0.001, 30 neurons in the first hidden layer, 10 neurons in the second hidden layer, and a Mean Squared Error (MSE) of 0.0487. Both methods demonstrated high classification accuracy, with ANN achieving 94% and LDA showing variable performance across oil types.