QBoost Regression on Preprocessed Small Data Sets Enables Optimization of Electrochromic Ink Compositions Containing Water-Dispersed WO ₃ Nanoparticles

The development of printable smart windows requires optimizing ink compositions that incorporate electrochromic nanoparticles to meet multiple, often conflicting, performance criteria. Although ink composition is the primary parameter we intentionally vary, such variation is linked to changes in several interconnected factors that cannot be directly controlled or fully observed; these can influence how effectively the nanoscale electrochromic particles exhibit their intrinsic properties. Emerging data-driven approaches offer a promising route for optimizing ink compositions, but their effectiveness decreases under information-sparse conditions, in which important factors influencing performance are lacking in the explicit inputs, thus requiring large sample sizes to statistically recover the missing relationships. Herein, we propose a data-driven framework that remains effective under small-data set conditions and demonstrate its applicability through the optimization of ink compositions containing tungsten oxide nanoparticles, a well-known electrochromic material, even with fewer than 15 experimental samples. By integrating the QBoost model─an annealing-based boosting model─with a carefully preprocessed data set, we achieved high predictive accuracy (R2 ≈ 0.9) for key electrochromic property indicators, including coloration efficiency, optical contrast, and response time. The trained model further suggests specific ink compositions expected to achieve balanced improvements across multiple electrochromic properties, offering practical guidance for subsequent experimentation.