JMIR Publications

ABSTRACT

Background:

Despite the increasing use of machine learning (ML) in clinical research, the early stages of data preparation—especially for structured clinical data—often receive limited methodological scrutiny. These datasets typically contain missing values, complex categorical variables, and imbalanced class distributions, all of which complicate downstream model development and interpretation.

Objective:

This study introduces a structured preprocessing framework designed to address common challenges in medical tabular data and to assess how preprocessing choices affect the stability and transferability of predictive models across settings.

Methods:

We constructed a modular workflow comprising three components. First, preprocessing strategies included imputation for missing data, three types of categorical encoding (One-Hot, Frequency, Target), and resampling approaches for class imbalance (SMOTE, ROSE). Second, six classification algorithms were used to evaluate performance patterns: Logistic Regression, Decision Tree, Random Forest, XGBoost, CatBoost, and LightGBM. Third, external validation was conducted across two datasets with distinct data-generating mechanisms: an ESRD patient registry (n=412) and the population-based BRFSS 2015 survey.

Results:

One-Hot Encoding in combination with ROSE yielded the most consistent performance improvements in terms of AUC (0.940) and accuracy (0.932), particularly for classifiers sensitive to class distribution. Notably, ROSE enhanced sensitivity without substantially distorting the original data structure. Feature importance rankings also contributed to model interpretability, and performance trends were largely reproducible in external validation.

Conclusions:

Our findings suggest that preprocessing decisions—often treated as ancillary—play a central role in shaping model outcomes, especially in high-variance clinical datasets. The proposed framework offers a reproducible and adaptable tool for aligning data preparation with the unique demands of healthcare prediction tasks, and may serve as a foundation for future efforts to standardize preprocessing in clinical ML workflows.