Prediction of type 2 diabetes based on feature augmentation and Morlet wavelet assisted deep learning network with FFT overlap and add convolution

The enormous development in technology makes diagnosing easier in the medical field under various existing approaches. Even though the approach possesses some disadvantages like lowering disease treatment costs, research showed that network features, which are important in decision-making but have a low accuracy value, were critical. A new deep-learning technique for diabetes detection is proposed in this research that resolves the challenges in the existing. The approach combines adversarial variational auto-encoder (AVAE) for data/feature augmentation with a Morlet wavelet-assisted deep learning network featuring fast Fourier transform (FFT) overlap and add convolution (MW-FFT-OAconv) to enhance classification accuracy. A novel optimiser, the weighted mean of vectors (WMOV), is introduced to acquire the weight parameters of the MW-FFT-OAconv network. Experimental results evaluated using statistical measures such as accuracy, F1-score, precision, true negative rate (TNR), true positive rate (TPR), classifier error percentage (CEP), and Mathew coefficient correlation (MCC), demonstrate the effectiveness of the proposed approach. Compared to previous machine learning prediction models (random forest, naive Bayes, and decision tree), the proposed technique achieves an accuracy level of 98.44% in predicting type 2 diabetes.