BiLSTM deep neural network model for imbalanced medical data of IoT systems

Health informatics is one of the most developed field in recent time. Computational Intelligence is among the most influential factors that may help to improve patient oriented and secure decision support model. In this article we present a model of IoT system, which combines BiLSTM deep learning with Decision Tree model and data balancing strategy used to help in automated diagnosis support. Presented solution include experimental series of data preprocessing using well established balancing algorithms with custom parameters and modifications in order to best prepare the data for the network training. Such algorithms are ADASYN, SMOTE-Tomek, etc. The system helps to evaluate questionnaires and securely exchange documents between patient and corresponding medical team. From the level of system patient and doctors are able to see automated diagnosis provided by deep learning model. The model gives an important advance to help patients faster. Results show that proposed BiLSTM deep learning with decision tree mode detects diseases from questionnaires with accuracy above 96%, precision above 88% and recall above 96% which proves efficiency of our proposed model.