Application of Feed Forward and Convolutional Neural Network Models in Diabetes Detection and Prediction Using Classification Algorithms

This study investigates the use of feedforward and convolutional neural networks for diabetes detection and prediction through classification algorithms. It utilizes a dataset of 495 records from various medical centers in Nigeria, which includes factors such as age, cholesterol levels, blood glucose levels, body mass index, physical activity, family history, and alcohol consumption. The data was split into 70% for training and 30% for testing, employing a multi-layer perceptron (MLP) classifier with Python and TensorFlow Keras. The MLP achieved an accuracy of 70%, while the convolutional neural network reached an accuracy of 71% after 500 epochs. The findings also indicated that 31.52% of alcohol consumers and 68.48% of non-consumers were affected by diabetes, highlighting alcohol consumption as a significant risk factor. The MLP recorded a precision of 49%, recall of 70%, and F1-score of 58% while the convolutional network had a loss of 0.5554, validation accuracy of 58% and validation loss of 1.0068.