The combined multilayer perceptron and logistic regression (MLP-LR) method better predicted the spread of Hyphantria cunea (Lepidoptera: Erebidae)

Abstract Hyphantria cunea (Lepidoptera: Erebidae) is one of the pests that pose a serious threat to forest and agronomic crops in China. Its spread is influenced by various factors, including environmental factors and anthropogenic factors, and the available data on pest spread and the influencing factor has nonlinear relationship. Additionally, the collection of pest data is often constrained, resulting in small datasets, a lack of long-term time series data, and issues such as missing data and anomalies. Traditional model-driven approaches have limitations in handling nonlinear relationships and high-dimensional data, while data-driven methods often lack interpretability and are prone to overfitting, ultimately leading to insufficient prediction accuracy. Therefore, this paper proposes the MLP-LR method, which combines logistic regression (LR) with a multilayer perceptron (MLP) to overcome these limitations. The model also used the Bayesian adaptive lasso method to select important influencing factors, that further improved the prediction accuracy. Based on H. cunea occurrence data in China, the current study demonstrated the stability and accuracy of the MLP-LR model on small datasets. The results showed that compared to traditional LR models and MLP independently, MLP-LR performs better in predicting the spread of H. cunea, effectively addressing the shortcomings of traditional methods. This study provides a new tool and perspective for forecasting and early warning of H. cunea outbreaks, offering important references for future research and its applications in the field.