Predicting the Activity Level of the Great Gerbil ( Rhombomys opimus ) via Machine Learning

TL;DR

This paper introduces a machine learning model to predict the activity of great gerbils, a pest species, to help manage their population and protect ecosystems.

Contribution

A novel PSO-ELM model is proposed for predicting the activity level of Rhombomys opimus with higher accuracy than traditional methods.

Findings

The PSO-ELM model achieved 91.67% accuracy in predicting fall activity levels of R. opimus.

Principal component analysis reduced data dimensionality from 92 to six components, improving model performance.

The PSO-ELM model outperformed the back propagation model in convergence and prediction accuracy.

Abstract

The great gerbil (
Rhombomys opimus
) is a pest rodent that is widely distributed in Eurasia, and assessing its outbreak risk and instituting timely population control are very important for protecting vegetation and human health. Because traditional assessment methods are difficult to monitor and cannot effectively predict the population growth trend of
R. opimus
, an
R. opimus
activity prediction model was constructed using the particle swarm optimization algorithm‐extreme learning machine (PSO‐ELM). First, data for 13 factors influencing
R. opimus
growth, such as those related to the environment, vegetation, and activity in the previous year, at 46
R. opimus
monitoring sites in China from 2020 to 2022 were selected. Second, principal component analysis was used to reduce the dimensionality of the 92 sets of collected data to six principal components, thus eliminating the correlation between the indicators. Third, after dimensionality reduction, the data were divided into a training set (80 sets of data) and a test set (12 sets of data) for model training and simulation, and the prediction results of the PSO‐ELM model and back propagation model were compared. The simulation results revealed that the PSO‐ELM model has a stronger convergence ability and higher prediction accuracy for the activity level of
R. opimus
in fall (91.67%). In this study, a new method is provided for surveying pest rodents. The proposed method provides an auxiliary means of managing
R. opimus
. We will continue to improve the sample data in future work to obtain more accurate predictions.

The particle swarm optimization algorithm–extreme learning machine was utilized to develop the hazard model for the great gerbil.

Figures