Improved Automatic Feature Selection Approach for Health Risk Prediction

With the recent advances in machines & technology, there is revolution in healthcare industry also. Thus, it gave rise to the concept of Electronic Health Record (EHR) which stores patients demographics, lab tests & results, medical history, habits etc. in electronic form. EHR is voluminous data which is difficult to store, maintain or alter. To provide extended life for people, health risk prediction model using this EHR is formulated in this work. Feature Selection is used to select only associated or relevant data from the dataset. Logistic Regression is used with improved loss functionality parameter which increases the accuracy, response time and performance of the system. Representation Learning enables formation of feature vector of the selected features thus calculating their scores. Further, risk prediction is performed by the neural network model. Deep Neural Network (DNN) is used with many hidden layers containing activation functions. Transfer learning is used to avoid re-training of the whole system every time new data enters the model. Dataset used here is of hypertension. EHR dataset is also synthetically created for analysis.