Utilizing machine learning for the assessment of mosquito repellent effectiveness and decision support in product selection

Abstract

The significance of effective insect repellents is underscored by the substantial threat posed by mosquito-borne diseases to global health. Machine learning (ML) algorithms have garnered attention as a potential approach for classifying insect repellents and evaluating their effectiveness. In this study, we explore the application of decision tree (DT), random forest (RF), logistic regression (LR), K-Nearest Neighbor (KNN), Naive Bayes (NB), AdaBoost (AB), Multilayer Perceptron (MLP), and Support Vector Machine (SVM) algorithms for categorizing insect repellents. ML techniques are employed to classify mosquito repellents by training various ML classification models using a labeled dataset in which each repellent is assigned a specific effectiveness level. These algorithms can identify patterns and determine the efficacy of a particular repellent by analyzing a range of features such as chemical composition, concentration, duration of protection, and environmental conditions. E-Nose technology offers several advantages, including high sensitivity, selectivity, rapid analysis, non-invasiveness, portability, extensive data analysis capabilities, cost-effectiveness, and versatility. In this research, we propose the use of a non-selective gas sensor array comprising eight MQ series elements, combined with ML analytics, to distinguish the type, concentration, and application of four different types of mosquito repellents commonly used in real-world scenarios. The sensor node prototype was exposed to each of the potential mosquito repellents for 15 to 20 minutes individually to create the experimental dataset. Subsequently, various ML models were trained using this dataset to accurately classify unknown data samples. SVM and MLP models achieved model accuracy scores of 98.01% and 98.10%, respectively, while DT, RF, and KNN models achieved training accuracies of 98.99%, 98.98%, and 98.61%, respectively. Classification performance error was also assessed using Mean Absolute Error (MAE), R2 Score, and Mean Squared Error (MSE). © International Journal of Sustainable Building Technology and Urban Development.