Enhancing the Accuracy of Diabetes Prediction Using Feedforward Neural Networks: Strategies for Improved Recall and Generalization

Abstract

This study explores the development and evaluation of a neural network model for predicting diabetes based on clinical data. The model was built using the Keras API with TensorFlow backend. Key steps included data preprocessing, such as feature scaling with `StandardScaler` and splitting the data into training and testing sets. The neural network architecture consisted of an input layer, two hidden layers with ReLU activation functions, and an output layer with a sigmoid activation function, optimized using the Adam optimizer and binary crossentropy loss function.

The model was trained over 50 epochs with a batch size of 10, incorporating a validation split of 20% to monitor performance on unseen data during training. The results demonstrated a high accuracy of approximately 97% on the test set, indicating the model's efficacy in predicting diabetes. Further analysis using a confusion matrix revealed a high count of true positives and true negatives, alongside minimal false positives and false negatives, confirming the model's robustness. These findings suggest that neural networks can be effectively employed for diabetes prediction, offering significant potential for integration into clinical decision support systems. However, further validation with larger and more diverse datasets, alongside considerations for data imbalance and model interpretability, is recommended to ensure generalizability and practical application in real-world healthcare settings.