Early Prediction of Health Disease for Long - Short Term Memory using AdaGrad Model

Diagnoses codes can be advantageous to decision support systems, which try to help coders by recommending a pertinent subset of probable codes to choose from. These systems are employed as a billing mechanism in the Electronic Health Record. Automatic International Classification of Diseases (ICD) code assignment is regarded as a challenging task in the field of multi-label classification due to the broad range of potential labels and length of patient records. The baseline for autonomous ICD code assignment presented in this research uses Long Short Term Memory (LSTM) in conjunction with an AdaGrad model for gradient-based optimization. As the parameters are linked to often occurring features, the model adjusts the learning rate to the parameters that execute smaller updates with low learning rates. With high learning rates for parameters connected to infrequent features, the larger updates improve. The proposed method obtained precision of 99.0% and recall of 90.8 % that was better compared to existing ANN models that obtained precision of 71.2% and recall of 66.8%.