Development of a Predictive Model for the Classification of the Risk of Hypertension Using Deep Learning

Information systems are now offering high-quality services as a result of the level of advancements that has been made on the Internet, mobile devices, and all other ICT infrastructure. Due to this, modern healthcare system has become more reliable providing innovative and effective means of reducing the burden of work on the part of the medical experts especially in a sub-Saharan country like Nigeria. This study aims to apply deep learning models to the classification of the risk of hypertension. The study identified and collected relevant data, simulated and validated the predictive model.
Relevant data containing information about the features that are associated with the risk of hypertension was collected from an online public repository by Kaggle. Feature selection methods were adopted for the identification of features that are more important for the detection of hypertensions from the dataset collected. The predictive model for the detection of hypertensions was formulated using a deep neural network architecture based on the factors that are associated with the detection of hypertensions. The simulation of the predictive model was achieved using the Google CoLaboratory, a Python jupyter notebook provided by Gmail based on the hold-one-out methods which adopted the training data for model building and a test data for model validation. The validation of the predictive model was achieved by assessing the performance of the predictive model using the test data based on a number of performance evaluation metrics, namely: accuracy, recall, precision and f1-score.
The result of this study was able to investigate the impact of relevant factors on the performance of classification of the risk of hypertension. The study revealed that the selection of relevant features can improve the performance of the detection of hypertension. The study showed that by selecting relevant factors from an initial number of initially identified factors, it is possible to reduce the space occupied by the data, reduce the complexity of the predictive model and to reduce the time taken for computing systems to process the model.
The study concluded that by focusing on the relevant factors, it is possible for physicians to avoid untimely deaths due to unforeseen circumstances as along as patients pay more attention to those factors. The study concluded that the use of feature selection can improve the performance of predictive models developed using deep learning and not just those built using machine learning algorithms.