Application of Supervised Machine Learning Models for the Identification of the Anxiolytic-like Effect Produced by Progesterone in Wistar Rats

Abstract

Machine learning is widely used to create mathematical models that explain or predict events based on previous observations. Within the most used algorithms are the naive Bayesian classifier, K- nearest neighbors or vector support machines. An area of potential application is behavioral pharmacology, that evaluates the behavior of experimental subjects injected with different substances to identify beneficial or toxic effects. Present study, classical statistical and machine learning techniques were used to evaluate the effect of progesterone (0.5 and 2 mg / kg) in the raised arms and open field maze. The results were compared between both data analysis approaches, identifying an anxiolytic-like effect of the 2 mg / kg dose of progesterone, similar to that produced by diazepam. The results of the analysis using classical statistical techniques show an anxiolytic-like effect of progesterone at a dose of 2 mg / kg. Consistently the machine learning techniques identified this effect, and further allowed generating predictive models with a reduced number of variables. This enabled automatically identify the variables that provide more information to differentiate the experimental groups.