Privacy-Preserving Hybrid GA-LSTM Ensemble for Typhoid Detection Using Optimised Clinical Feature Selection

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Computational Biology

Background:

• Typhoid fever presents a significant public health burden in low-income nations.
• Accurate diagnosis is challenging due to overlapping symptoms and unreliable conventional methods.
• Existing diagnostic tools often lack efficiency and patient privacy safeguards.

Purpose of the Study:

• To develop an automated, reliable, and privacy-preserving diagnostic framework for typhoid fever.
• To leverage clinical data for enhanced typhoid fever detection.
• To address limitations in current diagnostic procedures in resource-constrained environments.

Main Methods:

• A hybrid framework integrating genetic algorithm (GA)-based feature selection and a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) deep learning classifier.
• Federated learning with the Federated Averaging (FedAvg) algorithm for collaborative model training without raw data sharing.
• GA was employed to identify the most informative clinical features, reducing redundancy and computational load.

Main Results:

• The proposed framework achieved 92% accuracy with a strong F1-score and satisfactory sensitivity.
• The model demonstrated reduced memory requirements and shorter training times compared to using the full feature set.
• The approach maintained balanced performance even with class imbalance in the data.

Conclusions:

• The integration of evolutionary feature selection, deep sequential learning, and federated training offers an effective, privacy-aware solution for typhoid fever diagnosis.
• This framework is well-suited for clinical settings with limited data access and computational resources.
• The study highlights the potential of AI-driven solutions in improving infectious disease diagnostics in underserved regions.