A Technique Based on Ensemble Machine Learning for The Analysis of Electronic Nose Signals

Numerous applications have been documented for electronic nose (e-nose) devices, which are valued for their speed, affordability, and lack of invasiveness. The e-nose scheme is a useful instrument for analyzing volatile chemicals as a distinct 'fingerprint,' especially in the meat industry. The pattern identification algorithm's capacity to decode e-nose signals is crucial to the e-nose system's effectiveness in a wide variety of uses. However, positive results using ensemble approaches have been reported in a number of data types. In this study, an ensemble learning strategy is proposed for e-nose signal analysis, particularly for evaluating beef quality. Not only are learning algorithms and sensor array optimization using ensemble approaches. Ensemble FSA is created by combining three filter-based feature selection algorithms (FSAs)—reliefF, chi-square, and Gini index—for sensor array optimization. When using a single FSA on a collection of homogenous e-nose data for beef quality monitoring, it is common for the results to be inconsistent or unexpected. In addition, multi-class regression and categorization problems are handled by using ensemble learning techniques. We use bagging and boosting algorithms, shown here by Random Forest (RF) and Adaboost. Both a support vector machine (SVM) and a decision tree are used as independent learners, and their results are compared to those of the ensemble. Experiments show that our ensemble method is effective and generalizable for e-nose signal processing. Optimized sensor combination using filter-based FSA consistently outperforms previous methods in both classification and regression. Additionally, despite using a reduced sensor complement, Adaboost as a boosting algorithm yields the highest quality prediction.