A machine learning-based model for predicting the color of liquid formulation placebos and analyzing their visual...

High-quality randomized, placebo-controlled trials are critical for clinical drug trials, but preparing color‑matched placebos for liquid formulations remains difficult due to the color complexity of liquid formulations and the inefficiency of conventional matching methods. Machine learning offers an effective approach to modeling the complex non-linear relationships among multiple colorants at varying concentrations, enabling prediction of color formulations. This study employed optimal Latin hypercube sampling to generate 253 simulated samples with five colorants. After color correction using a correction matrix, ten color space values were extracted for model fitting. Among five models, random forest (RF) performed best, with a mean absolute error of 0.09, a root mean square error of 0.15, and an R² value of 0.69. Shapley additive explanations analysis identified L*, G, gray, and V as key variables influencing the prediction of placebo color formulations. Validation against eight target solutions produced average color differences of ΔE_ab = 3.35 and ΔE₀₀ = 2.20, and a mean visual similarity score of 9.1 by human assessors, indicating a relatively high prediction accuracy. This study developed an RF‑based strategy for predicting the color of placebo in liquid formulations, enabling visual color matching with test drugs in clinical trials.