A Framework for Unsupervised Profiling of Malaria Vectors' Insecticide Resistance Using Machine Learning Technique

Ashuza Kuderha,Wisdom Adingo,Bruno Chikere,Mugisho Kulimushi,Kala Jules

DOI: https://doi.org/10.1089/vbz.2023.0112

Abstract:Background: There is a need to identify different insecticide resistance profiles that represent circumscription-encapsulation of knowledge about malaria vectors' insecticide resistance to increase our understanding of malaria vectors' insecticide resistance dynamics. Methods: Data used in this study are part of the aggregation of over 20,000 mosquito collections done between 1957 and 2018. We applied two data preprocessing steps. We developed three clustering machine learning models based on the K-means algorithm with three selected datasets. The elbow method was used to fine-tune the hyperparameters. We used the silhouette score to assess the clustering results produced by each of the three models. The proposed framework incorporates continuous learning, allowing the machine learning models to learn continuously. Results: For the first model, the optimal number of clusters (profiles) k was 17. For the second model, we found four profiles. For the third model, the optimal number of profiles was 7. Discussion: We found that the insecticide resistance profiles have dynamic resistance levels with respect to the insecticide component, species component, location component, and time component. This profiling task provided knowledge about the evolution of malaria vectors' insecticide resistance in the African continent by encapsulating the information on the complex interaction between the different dimensions of malaria vectors' insecticide resistance into different profiles. Policy makers can use the knowledge about the different profiles found from the analysis of available insecticide resistance monitoring data (through profiling) by using our proposed approach to set up malaria vector control strategies that consider the locations, species present in those locations, and potentially efficient insecticides.