Evaluating Non-Invasive Haemoglobin Estimation: A Comparison Analysis of Machine Learning Frameworks on Simulated Healthcare Data

Conventional haemoglobin measurement methods, like laboratory blood tests, are invasive and may cause discomfort for patients. This study uses simulated healthcare data to assess different machine learning models for predicting non-invasive hemoglobin estimation (NIHE). Ten thousand samples make up the dataset, which includes information on the patient's age, gender, skin tone, blood pressure, heart rate, SpO2, real hemoglobin levels, device used, and ambient conditions. With an emphasis on comparing model performance, the objective is to predict NIHE based on these features. Testing was done on five different machine learning models: K-Nearest Neighbors Regressor, Support Vector Regressor, Gradient Boosting Regressor, Random Forest Regressor, and Linear Regression. Three performance indicators were used to evaluate the models: R-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), Mean Squared Error (MSE), and Mean Absolute Error (MAE). With an MAE of 1.264, MSE of 2.491, and R2 of 0.197, the results show that the Gradient Boosting Regressor performed the best. With an MAE of 1.265, MSE of 2.499, and R2 of 0.194, Linear Regression trailed closely behind. With an MAE of 1.386, MSE of 2.991, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.036, on the other hand, the K-Nearest Neighbors Regressor performed the worst. With R2 values of 0.155 and 0.154, respectively, the Random Forest Regressor and Support Vector Regressor likewise fared somewhat worse than the best models. The findings imply that although sophisticated ensemble techniques such as Gradient Boosting offer marginal advantages over more basic models, the accuracy of noninvasive hemoglobin level prediction is only moderately improved by any of the examined methods. This study emphasizes how difficult it is to create reliable prediction models for use in healthcare applications and proposes that feature engineering or additional model exploration and refining may be required to improve predicting.