Developing Accurate Predictive Models for Phase Angle Estimation in Adult Individuals of Southern Cuba Region using Classification and Regression Learner Methods from Bioimpedance Measurements

Abstract Objective: The combined use of bioimpedance and machine learning is essential for accurately estimating the health status of individuals. This work is aimed to develop a machine learning predictions of phase angle derived from bioimpedance measurements of adult-healthy volunteers from Santiago de Cuba, Cuba. Methods: We conducted a pilot study in Santiago de Cuba, Cuba with a total of 2848 volunteers between the ages of 19 and 96. The study included individuals with asthma, diabetics, hypertension, ischemic heart disease and with no apparent illness. Bioimpedance parameters were measured using a standard tetrapolar whole-body configuration. A classification and regression model was used to predict the phase angle. Results: The classification model identified relevant bioparameters for predicting the phase angle. Total body water showed the greatest importance at approximately 94%, followed by the corrected capacitance at ~18% and the basal energy costs at ~17%. The predicted phase angle values were consistent with previous reports of other pathologies, where higher phase angle values indicate better health and men have higher values than women. The model suggests that Xc, TBWkolt, and GBE can be used as alternative indicators for estimating health status. Furthermore, a previously unreported relationship between the metabolic basal index and phase angle was proposed with possible implications for health status. Conclusion: The classification, phase angle and body mass index predictive models are of the great importance to assist the exploration of partial heath status with great accuracy. The results reveal that the predictor models can be used for estimating the phase angle and intrinsically the health status of adult patients with helpful diagnosis assistance in Santiago de Cuba, Cuba.