A Procedure to Improve Binary Classification Models and Categorize Features: The Case of the Distribution of Three Mosquito Species in Morocco

Abstract

Modeling biological datasets represents an essential step in processing and exploiting biological information.Selecting features and improving modeling quality are critical in building a high-performance predictive model.In this article, we have presented and applied a novel approach to select features and to improve the modeling quality using the presence/absence data of three mosquito species in Morocco.This approach uses a recursive search of feature subsets conditioned on improving the modeling quality compared to an initially chosen solution.It has led to a significant improvement in the modeling quality compared to another study carried out on the same dataset, where the accuracy of the models improved with a range varying between 0.062 and 0.198.The relevance of this approach also extends to the search for solutions that achieve the same performance with different subsets, known as multiple solutions.These solutions demonstrate that various combinations of explanatory features can explain the target feature, leading to categorizing them according to their impact on the modeling.This work has provided a good explanation of the distribution of mosquito species thanks to the improved modeling quality, opening up the possibility of having relevant solutions and discovering new explanatory modes for the features.