MRC: A Hybrid Relevance and Correlation Metric for Indicator Selection with Application to SINASC Data

Variable selection is a critical step in data analysis and modeling, especially in domains with high-dimensional datasets such as healthcare. Addressing this challenge is particularly crucial in digital governance, where robust indicator prioritization underpins evidence-based policymaking. Traditional approaches often rely solely on statistical correlation or feature importance derived from predictive models, potentially overlooking semantic relevance. In this paper, we introduce the Metric of Relevance and Correlation (MRC), a novel hybrid approach that combines semantic relevance, statistical correlation, and predictive impact to identify the most relevant variables for a target outcome. We apply MRC to the Brazilian Live Birth Information System (SINASC) data to identify key factors associated with the 5-minute APGAR score, an important indicator of newborn health. Our results show that MRC provides a more stable variable ranking compared to traditional correlationbased and random forest-based methods, with superior performance across multiple stability metrics. MRC successfully identifies clinically relevant variables while discovering non-obvious relationships that traditional methods might miss. This approach offers a more comprehensive framework for variable selection, applicable across various domains requiring robust feature prioritization, and is particularly valuable for digital government initiatives in aiding the formulation of more effective public policies.