|
چکیده
|
CHD is one of the biggest challenges in health care due to its nature and specifically due to the necessity of early and correct diagnosis. The classification approaches main issue is that the traditional diagnostic methods are not very useful in handling the class imbalance issue and in achieving the high predictive accuracy. To solve these challenges, this thesis introduces an enhanced prediction model based on GANs and optimised CNNs. The proposed method starts with data pre-processing to handle a disproportionate distribution of normal and CHD records. This entails data cleansing and formatting, normalization and utilization of TGAN to synthesize, additional samples of CHD to balance the dataset. The SOC is followed by three convolution blocks, a dropout layer the global average pooling layer a fully connected layer with a softmax activation function for classification. To optimize the CNN’s performance, the Emperor Penguin Colony Optimization (EPCO) algorithm is employed to fine-tune four critical hyperparameters: filter size, the number of filters, learning rate and the batch size. It is shown that the proposed method yields high level of accuracy in rate, reaching 99.46% for the training data set and 99.45% for the testing data set. Such outcomes confirm stability, as well as high accuracy of the model in terms of CHD prognosis. When TGAN is used for data augmentation and EPCO for hyperparameter optimization, the performance improvement of the model for handling class imbalance considerably improves. The strength of being able to use this method is that it increases the predictive accuracy and that handling of imbalanced data sets. This integrated solution not only improves the state of art of predicting CHD but also lays out a blueprint which can be fine-tuned for other medical diagnostic problems. The best future directions for this kind of study include assessing the model on a range of datasets, exploring further data enhancement and optimization methodologies, incorporating this model to other diagnostic systems, and optimizing the interpretability, and real-time capability of the model. These efforts will build on developing the model over the following steps in an effort to increase the model’s usefulness in the early identification of congenital heart disease and its treatment.
|