Modelling lymphatic filariasis dynamics using Levenberg–Marquardt algorithm-artificial neural networks

Abstract

This paper introduces a hybrid Levenberg–Marquard-Artificial Neural Network (LMA-ANN) framework for modeling the complex transmission dynamics of lymphatic filariasis (LF), a debilitating vector-borne neglected tropical disease. The methodology addresses key challenges in data-driven epidemiological forecasting by combining the fast convergence properties of the Levenberg–Marquardt optimization algorithm with the universal function approximation capability of neural networks. We evaluate the proposed framework against four established neural architectures such as Multilayer Perceptron (MLP), Fully Connected Neural Network (FCNN), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN) using both pristine and Gaussian noise-augmented synthetic datasets generated from a compartmental epidemiological model solved with a high-fidelity Runge–Kutta method. Results demonstrate that the LMA-ANN achieves superior predictive accuracy, with the lowest error metrics of $MAE=0.029,RMSE=0.039,MSE=0.0015$ and highest coefficient of determination of $R^2=0.990$ on noise-augmented data, while maintaining computational efficiency with the shortest training of $87.4s$ and inference of $2.9\text{ms}$ times. Crucially, the CNN and RNN architectures exhibited worst performance degradation on the noise-augmented dataset, yielding negative $R^2$ values of $-0.15$ and $-0.42$ respectively, indicating predictions worse than a simple mean model. This highlights a critical limitation of complex architectures when trained on limited, noisy epidemiological data. The study provides two principal contributions: (1) a robust, computationally efficient LMA-ANN framework that accurately captures LF dynamics under realistic data constraints, and (2) evidence-based guidance for model selection in epidemiological applications, emphasizing that architectural complexity must be carefully matched with data quality and quantity. These findings advance computational methods for infectious disease modeling and offer a generalizable tool for public health decision-making in resource-limited settings.