Using Time-Series Models for Heart Rate Prediction

Introduction:

Heart rate prediction plays a pivotal role in healthcare by providing essential insights into a patient's overall well-being. In this blog post, we delve into the significance of heart rate prediction and conduct a comprehensive analysis of three time-series models: ARIMA, SARIMAX, and Exponential Smoothing.

Dataset Description:

Our dataset, derived from a Lifetouch device, encompasses heart rate and respiration rate data, along with SpO2 and pulse data from an oximeter. With approximately four hours of data collected at 1-minute intervals, the dataset consists of 226 entries and 5 columns.

Data Pre-Processing:

In the heart rate prediction task, meticulous data pre-processing was conducted. This included data cleaning, handling missing values, data normalization, and addressing outliers.

Comparison of the Models:

Before delving into model selection, we ensured the dataset's stationarity. Subsequently, we compared the performance of our three chosen models using various metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), R-squared, AIC, and BIC. Our rigorous analysis unveiled that the Exponential Smoothing model, particularly the triple exponential smoothing variant, exhibited the most accurate predictions and excelled in all evaluation metrics.

Analyzing the Results and Choosing the Best Model:

The selection of the most suitable model is of paramount importance in machine learning tasks. Our extensive analysis unequivocally identifies the Exponential Smoothing model as the optimal choice for heart rate prediction. This model leverages a combination of trend, seasonality, and error components within the dataset to ensure precise heart rate predictions.

Conclusion:

In conclusion, our in-depth exploration of ARIMA, SARIMAX, and Exponential Smoothing models reveals that the Exponential Smoothing model, with its triple exponential smoothing technique, stands out as the ideal choice for heart rate prediction. This model holds immense potential for real-time applications, enabling healthcare professionals to monitor patient health effectively and provide valuable insights.

References:

1. "Time Series Analysis and Its Applications: With R Examples" by Robert H. Shumway and David S. Stoffer

2. "Forecasting: Principles and Practice" by Rob J Hyndman and George Athanasopoulos

3. "Introduction to Time Series and Forecasting" by Peter J. Brockwell and Richard A. Davis

4. "Applied Time Series Analysis for Fisheries and Environmental Science" by Richard D. Methot, Jr.

5. "Time Series Analysis: Forecasting and Control" by George E.P. Box, Gwilym M. Jenkins, and Gregory C. Reinsel.

Code for reference: 

 https://github.com/yamini542/AppliedAI_Assignments/tree/main/Assignment_1_TimeSeries