Objectives:
The primary objective of the study is to classify the blood pressure (BP) levels using advanced machine learning (ML) techniques for predictive purposes. The study assesses the efficacy of the Naïve Bayes, AdaBoost, feedforward neural networks (FNNs), and long short-term memory (LSTM) algorithms over the conventional multinomial logistic model using standard performance evaluation metrics.
Methods:
The dataset comprised 15,000 entries obtained from the National Health Service, England, each containing eight variables. The variables include BP, age, weight, height, gender, smoking habit, alcohol consumption, and fitness level. The Naïve Bayes, AdaBoost, FNN, LSTM, and multinomial logistic models were employed in the study. Each model underwent training, testing, validation, and evaluation using suitable metrics such as accuracy, F1-Score, kappa statistics, sensitivity, specificity, and area under the curve score.
Results:
The FNN model gives the highest test accuracy of 89.47% and balanced performance, making it the most appropriate model for predicting BP levels. The LSTM model demonstrated strong proficiency in capturing temporal patterns. AdaBoost was highly effective for dealing with class imbalance, but Naïve Bayes was a dependable benchmark. The multinomial logistic model established a reliable and stable reference point. The results represented a notable improvement over previous research, which typically reported median accuracy rates in the 80–85% range.
Conclusion:
The study reveals that knowing an individual’s age, weight, height, gender, smoking habit, alcohol consumption, and fitness level is useful in predicting his/her BP level. Thus, the advanced ML algorithms demonstrate potential in accurately classifying BP levels and can aid in the prevention, detection, and management of hypertension.
