MpoxNet: Leveraging hybrid deep learning for enhanced monkeypox diagnosis and risk identification

Abstract

Monkeypox is a rare viral disease historically documented in Central and West Africa, though its recent emergence in global outbreaks has raised significant public health concerns. Accurate and timely diagnosis is crucial for effective containment. While existing research predominantly focuses on image-based diagnostics, the potential of tabular data—which is vital for symptom-driven screening and epidemiological analysis—remains largely underexplored. Furthermore, developing robust diagnostic models from tabular data is challenged by limited sample sizes and high data heterogeneity. To address this, we propose MpoxNet, a novel hybrid deep learning model that integrates Long Short-Term Memory (LSTM) networks with Multi-Layer Perceptrons (MLPs) to classify monkeypox cases and identify associated risk factors, particularly among HIV-positive individuals. We preprocessed publicly available Kaggle datasets by applying resampling techniques to mitigate class imbalance and conducted symptom correlation analysis to improve feature representation. Our experimental results demonstrate that MpoxNet achieved an accuracy of 65.35%, a precision of 65.04%, and a recall of 65.68% on Dataset D1, and an accuracy of 87.50%, a precision of 73.33%, and a recall of 100% on Dataset D2. For comparison, we evaluated traditional ensemble models, including AdaBoost, XGBoost, and Random Forest, which consistently outperformed baseline classifiers. These findings highlight the significant diagnostic value of tabular data and establish a foundation for deploying AI-driven, symptom-based tools to augment clinical decision-making and enhance public health surveillance strategies for monkeypox.