Comparing machine learning, deep learning, and reinforcement learning performance in Culex pipiens predictive modeling

Abstract

Several machine learning (ML) and deep learning (DL) methods have been used to predict the presence of species in classification problems. Another set of methods, called reinforcement learning (RL), has been used in training agents to perform various tasks, but not in predicting species distribution. Culex pipiens (Diptera: Culicidae), commonly known as the common house mosquito, is a globally distributed species prevalent in temperate and subtropical regions. They serve as a primary vector for West Nile Virus (WNV), a mosquito-borne pathogen that affects humans and other animals. The study objective is to compare the performance of logistic regression, random forest classifier, deep neural networks, and the RL methods, including Q-learning, deep Q-network (DQN), REINFORCE, and Actor-Critic, in predicting the historical presence of C. pipiens through their potential geographic distribution in the USA. The comparison showed similar performance across approaches, with reinforcement learning methods like DQN and REINFORCE showing effective performance using fewer features, making them as great prediction tools for changing environments or situations with limited resources. Moreover, the results revealed that altitude and annual precipitation were the most important bioclimatic variables predicting the historical presence of C. pipiens.