A Predictive Analysis of Heart Rates using Machine Learning Techniques

  • Matthew Oyeleye

Student thesis: Master's Thesis


Heart illness is one of the notable diseases of public health importance globally, induced by a reduced heart rate. Hence, early monitoring and screening of the heart wellness would help to detect anomalies in the heart rate (HR) in advance thus; managing irregularities in the heart function at the outset. The increase utilization of progressive technologies such as artificial intelligence (AI), Internet of Things (IoTs) and, wearable monitoring systems in health sectors has continue to be a significant part in the analysis of considerable health-based data for early disease detection, as well as accurate diagnosis, and to aid in the prognosis and evaluation of treatments. Therefore, analysing the effectiveness of data analytics and machine learning (ML) usage in the monitoring and prediction of heart rates making use of data generated by wearable device (e.g., accelerometer) is pivotal. Consequently, in this study, a number of efficient data-oriented models were looked into. These includes; the Autoregressive Integrated Moving Average (ARIMA), Linear Regressor (LR), SupportVector Regressor (SVR), K-nearestNeighbors Regressor (KNNR), DecisionTree Regressor (DTR) and, RandomFrorest Regressor (RFR) models, also, Long Short-Term Memory (LSTM) recurrent neural network algorithm, in analysing a time-series accelerometer generated univariant HR data from healthy individuals to forecast heart rates. The models were pipelined with a with a 3-fold cross validation (CV) GridSearchCV hyperparameter tunning to find the best estimator settings for the models’ development.

The models’ performances were evaluated on a recently generated data source with different time recording durations. The results of the experiments indicate that the ARIMA and the LR models can effectively predict heart rate in all recording time durations while the remaining models can perform best with above 1 minute recording duration. Thus, the experimental results established that the models are good tools in predicting and monitoring of more precise HR with the aid of accelerometer.
Date of Award29 Sep 2022
Original languageEnglish
SupervisorTianhua Chen (Main Supervisor) & Grigoris Antoniou (Co-Supervisor)

Cite this