Incorporating Machine Learning Techniques to Enhance Rodent Surveillance in Marginalized Urban Communities

TL;DR

The paper introduces a machine learning approach to improve rodent surveillance in urban areas by automating track plate analysis, making it faster and more cost-effective.

Contribution

A novel machine learning framework is proposed to automate rodent track plate interpretation, offering a cost-effective alternative to manual methods.

Findings

Machine learning methods like PCA and LM achieved results comparable to manual analysis of track plates.

The proposed approach reduces time and cost for rodent surveillance in resource-limited settings.

Dimensionality reduction techniques helped identify key patterns on track plates efficiently.

Abstract

Effective management of rodent pests necessitates efficient population surveillance. Many of the available methods currently used for estimating rodent populations are either costly or time‐intensive. Rodent trapping demands significant resources, while tracking plates (TP) require high technical expertise and weeks to months of dedicated effort to satisfactorily interpret the plates. Here, we propose integrating Machine Learning techniques to evaluate plates with signs of rodent marks and compare their accuracy with that of conventional human‐interpreted plates. We employed the Otsu method to transform plates from RGB color images to grayscale images, highlighting regions of interest. Subsequently, we applied a global threshold to create binary images, assigning values above a globally determined threshold as 1s and others as 0s. The original images were transformed into new versions with 25 small samples, highlighting regions of interest based on the binary images. We used dimensionality reduction methods to identify the fundamental structure of high‐dimensional data and determined the most important patterns of interest on the plates. Among the methods, Principal Component Analysis, Independent Component Analysis, and Legendre Moments methods were used to visualize patterns and conduct exploratory data analysis. The k‐nearest neighbors, a versatile and intuitive classification method relying on the similarity principle, predicted the feature vector of PCA, ICA, and LM (Lpq) results. Ultimately, results from PCA and LM compared favorably against the conventional labur‐intensive manual method, thus proffering those in the field of disease ecology a better alternative for conducting timely and cost‐effective rodent surveillance to monitor rodent distribution hotspots during rodent management programs. We propose a novel approach that could significantly enhance the protocols of rodent surveillance programs, particularly in Low‐ and Middle‐Income Countries, where expertise in interpreting TPs may be limited to enhance rodent surveillance evaluation and timely rodent management while contributing to the indirect control of rodent‐borne zoonoses.

Represent the in‐field design of the trackplates and analysis methods.

Figures