Deep hyper optimization approach for disease classification using artificial intelligence

Abstract

Disease classification using Artificial Intelligence (AI) is one of the emerging areas for medical professionals to diagnose the disease. There are common diseases like breast cancer, hepatitis, thyroid and heart attack are faced by most of the people that produce severe health problems. An Artificial Neural Network (ANN) is a part of AI used to identify the nonlinear relationship among the features for better prediction. However, there are some common problems like under fitting, overfitting, increased elapsed time and vanishing gradient that occur during the analysis and prediction which reduces the performance of the model. So, there is a need for complex structure recognition without overfitting and under fitting. The present study suggests the Deep Hyper Optimization (DHO) technique to reduce the elapsed time of execution. It is used to fine tune the weight, bias and identify the optimal number of hyper parameters in the hidden layer. As a second opinion to medical professionals, various state of art classification models are used such as Logistic Regression , Decision Tree , Random Forest , K Nearest Neighbor and Gradient Boosting algorithms. The performance of state of art models are observed and compared with Deep Hyper Optimization (DHO) technique. It chooses the best hyper parameter for classification based on the highest probability based on the divide and conquer approach . The proposed model is tested for four different datasets and the performance of model is observed based on the accuracy, elapsed time, precision, recall, and F1 score, FPR, FNR , TNR and Area under Curve (AUC). • A Triplet Feature Grouping (TFG) is applied to reduce the dimensionality based on Pearson correlation value. • Deep Hyper Optimization (DHO) uses the divide and conquer approach to select the optimal parameters from the given list. • This reduces the elapsed time and increase the parallelism to reduce the memory cycle time. • DHO and TFG increase the accuracy of all the dataset above 95% and reduce the elapsed time of the model.