Establishing proper trust between human workers and robots is crucial for ensuring safe and effective human-robot interaction in various industries, including construction. An accurate trust prediction facilitates timely feedback and interventions, helping workers calibrate their trust levels. While machine learning modeling personalization (i.e., tailoring models to individual characteristics) has garnered attention in the literature, the conventional approach of developing a personalized model for each individual is impractical in labor-intensive industries like construction. Such an approach compromises efficiency and leads to an accuracy-efficiency tradeoff. To address this gap, this study aims to investigate the tradeoff inherent in model personalization and identify a cost-effective solution to enhance trust prediction accuracy without compromising efficiency. The results suggested that a partial model personalization method can effectively balance this tradeoff. Moreover, the proposed feature-based partial personalization approach enables a cost effective trust prediction model development for the construction industry, demonstrating its broader applicability to other worker-related predictions in other settings. This study provides insights into the strategies to improve trust prediction accuracy while maintaining the efficiency of model development by considering the distinctiveness of the future construction industry.