Multihead Attention based Convolution Neural Network for Cardiovascular Disease Prediction and Classification

Cardiovascular disease is becoming major cause of increased mortality rate across the world from clinical reports of world health organization. According to world health organization, early diagnosis of the heart disease can largely mitigate the mortality rate. Thus numerous researches have been carried out in medical industries across the world towards early diagnosis of the diseases and classification of disease into stages on basis of the heart condition. Especially manual diagnosis of heart disease based on the clinical data was highly prone to prediction and classification error. Due to advancement in the computer vision and artificial intelligence techniques, it is possible to automate the detection and classification using machine learning architecture and deep learning architecture. However, curse of dimensionality and class imbalance issues been a persistent challenge to the machine learning architectures towards accurate prediction analysis. In this paper, those challenges were mitigated on designing a new architecture entitled multi class attention network which is highly efficient in handle large dimensional dataset. Initially dataset is preprocessed with missing value imputation approaches to handle missing value in the disease attributes. Next, preprocessed data is applied to Multiclass Attention Convolution Neural Network. Multiclass Attention Convolution Neural Network processed with attention mechanism in the convolution layer. In convolution layer, relevant feature is extracted using kernel function and it is organized in form of feature map. Attention mechanism is to select the disease specific feature on refining the feature map to generate the attention map. Further attention map is processed with attention weight on pooling layer to compute the feature attention. Those feature attention and its map is processed in the fully connected layer using batch normalization function, activation function before classifying the feature in the softmax function. Softmax function uses the naive bayes classifier to classify the disease feature into heart disease classes based on condition of heart. Finally, loss function using cross entropy and epoch value were set to eliminate the over fitting and class imbalance challenges. Experimental analysis of the proposed architecture is carried out using benchmark EHR dataset to evaluate the model accuracy and efficiency. On performance evaluation of model, it is proved that proposed model outperforms state of art approaches on accuracy metric. Proposed model produces accuracy of 98.6% respectively on cross fold validation through confusion matrix.