Bias and Generalizability Challenges in Machine Learning Models for Leptospirosis

This paper highlights the presence of bias in machine learning (ML) and deep learning (DL) models used in leptospirosis research. Of the 17 studies reviewed, only four showed low analysis bias, while the remaining 13 had issues such as data imbalance, small sample sizes, and geographic bias. These biases negatively affect model performance and limit generalizability. We explore the sources of these biases—such as feature selection, dataset composition, and model design—and discuss their impact on predictive accuracy. Finally, we propose strategies to reduce these biases, aiming to improve the fairness, reliability, and performance of ML/DL models in leptospirosis prediction and diagnosis.