A deep learning approach based on convolutional LSTM for detecting diabetes.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Biomedical Data Analysis

Background:

• Diabetes mellitus is a serious chronic condition requiring early detection for effective management.
• Conventional diabetes diagnosis methods can be time-consuming and complex.
• Automated systems are needed to efficiently classify diabetes using clinical and physical data.

Purpose of the Study:

• To develop and evaluate a novel diabetes classification model using Convolutional Long Short-term Memory (Conv-LSTM).
• To compare the performance of the Conv-LSTM model against Convolutional Neural Network (CNN), Traditional LSTM (T-LSTM), and CNN-LSTM models.
• To identify significant features for accurate diabetes classification.

Main Methods:

• A new diabetes classification model based on Conv-LSTM was developed.
• The Pima Indians Diabetes Database (PIDD) was utilized for model training and testing.
• The Boruta algorithm was employed for feature selection, identifying glucose, BMI, insulin, blood pressure, and age as key predictors.
• Hyperparameter optimization was performed using Grid Search.
• Model performance was evaluated using both standard train-test splits and cross-validation techniques.

Main Results:

• The Conv-LSTM model achieved the highest accuracy of 91.38% in initial experiments.
• Using cross-validation, the Conv-LSTM model reached a peak accuracy of 97.26%.
• The Conv-LSTM model demonstrated superior performance compared to CNN, T-LSTM, and CNN-LSTM models, as well as existing state-of-the-art methods.

Conclusions:

• The developed Conv-LSTM model offers a highly accurate and efficient automated approach for diabetes classification.
• Key clinical features like glucose, BMI, insulin, blood pressure, and age are crucial for precise diabetes detection.
• This novel model holds significant potential for improving early diabetes diagnosis and patient outcomes.