Influenza, dengue and common cold detection using LSTM with fully connected neural network and keywords selection

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Machine learning models offer a solution to reduce physician workload in disease detection.
• Existing research utilizes machine learning and deep learning for disease detection and clinical classification based on patient symptoms and vital signs.

Purpose of the Study:

• To develop and evaluate an improved machine learning model for disease detection.
• To enhance the accuracy of symptom-based disease detection by integrating diverse patient data.

Main Methods:

• A hybrid model was developed using Long Short-term Memory (LSTM) for processing text-based patient symptoms.
• A fully connected neural network was integrated to incorporate additional patient data: body temperature, age, gender, and month of care.
• A data preprocessing algorithm was enhanced with keyword selection to mitigate overfitting and reduce input complexity.

Main Results:

• The combined LSTM and fully connected neural network model demonstrated superior performance compared to the standalone LSTM model.
• The implemented keyword selection method significantly improved the overall performance of the machine learning model.
• The hybrid approach effectively utilized both textual symptom data and structured patient information.

Conclusions:

• The hybrid machine learning model integrating LSTM and fully connected neural networks shows promise for accurate disease detection.
• Keyword selection is an effective strategy for enhancing the performance and preventing overfitting in symptom-based diagnostic models.
• This approach has the potential to support clinical decision-making and improve healthcare efficiency.