Classifying smoking status using linear and non-linear models based on clinical health records

Routine clinical chemistry data are widely collected in medical settings, yet their potential for lifestyle characterization using multivariate analysis remains underexplored. Leveraging these routinely available measurements could provide a cost-effective strategy for identifying lifestyle-related biochemical patterns at the population level. In this study, a range of linear and non-linear multivariate classification methods were evaluated to discriminate between smokers and non-smokers using 23 routine clinical chemistry measurements. Linear approaches included traditional partial least squares discriminant analysis (PLS-DA), PLS-DA with bootstrap resampling, and logistic regression (LR), while non-linear models comprised support vector machines (SVM) and random forest (RF). The results revealed differences between linear and non-linear classification strategies. Random forest showed comparatively better classification performance for the present dataset under the evaluated conditions, suggesting its ability to capture complex relationships within the biological data. Variable importance analysis highlighted the cholesterol ratio, total protein, potassium, and lactate dehydrogenase as relevant contributors to class discrimination, suggesting systemic metabolic and physiological differences between smokers and non-smokers. Overall, the findings demonstrate that routinely available clinical chemistry parameters, when coupled with appropriate multivariate analysis, can effectively capture smoking-related biochemical alterations. This study contributes to the fields of clinical chemometrics and data-driven healthcare by demonstrating that standard laboratory measurements can support lifestyle stratification, offering a practical and accessible complementary screening strategy prior to more detailed metabolomic investigations.