An ANN-BP based Multi-Alcohol Classification Model Through Different QCM Gas Sensors

Abstract

Alcohol content detection provides several advantages and is crucial in research encompassing product and business development. The findings of a sensor’s precise detection cannot be immediately classified; instead, they need to be processed further to identify the proper type of alcohol. A quartz crystal microbalance (QCM) is one among many sensors that can identify various types of alcohol. In order to identify numerous categories of alcohols, this paper presents an ANN-BP based model trained over 5 different datasets of QCM 3, QCM 6, QCM 7, QCM 10, and QCM 12. The ANN-BP model employs keras optimizer, one-hot encoding to display results and is trained via a fixed learning rate. In this article, five different forms of alcohol are used i.e., 1-octanol, 1-propanol, 2-butanol, and 1-isobutanol. The model is run through several epochs of training with a learning rate of 0.01 and the impact of loss and accuracy with respect to the number of epochs is discussed in detail. The performance analysis of the classification of the test set is carried out by plotting the confusion matrix as well as the metrics of precision, recall, accuracy, and f1 score. The model is also trained by alteration of different hyperparameters and analysis is carried out based on their results.