| Abstrakt: |
Diabetes is a globally dangerous disease characterized by decreased insulin levels and increased blood sugar which causes significant health risks and requires careful management. However, the class imbalance issue leads to a biased model resulting in poor prediction performance due to insufficient pattern learning. In this research, the Parametric Swish-based Recurrent Neural Network (PSRNN) is proposed to accurately and effectively predict diabetes. Initially, the data is gathered from the Pima Indian Diabetes Dataset (PIDD) and Frankfurt Hospital, Germany datasets to analyze the model's performance. Then, min-max normalization is used to normalize the data by scaling process which preserves the relative relationships among data points. Then, the Adaptive Synthetic Technique (ADASYN) is employed to balance the imbalanced data in diabetes prediction which improves the model's performance by generating synthetic minority class samples. At last, PSRNN is performed for diabetes prediction. The PSRNN achieves a better accuracy of 99.81% and 0.992 on the PIDD and Germany datasets, in comparison to the existing techniques like Support Vector Machine (SVM), Deep Convolutional Neural Networks (DCNN), and Artificial Neural Networks (ANN). [ABSTRACT FROM AUTHOR] |
