Machine Learning for Controlling the Particle Size Distribution of TiO₂ Nanoparticles in Optical Nanocomposites

To achieve precise control over the particle size distribution (PSD) of titanium oxide (TiO 2 ) nanoparticles (NPs) in optical nanocomposites, this study introduces a GBDT-RR (Gradient Boosted Decision Tree-Rosin-Rammler) framework to guide their synthesis. We first constructed a TiO 2 synthesis-PSD data set based on hydrodynamic diameter by combining Syn-ext Llama with manual collaboration. Based on this data set, we integrated a GBDT model for high-precision prediction with an RR model for concise PSD representation. Moreover, this framework was embedded into a Bayesian optimization (BO) algorithm (GBDT-RR with BO) to develop an efficient synthesis reverse design platform. Experimental validation confirmed the platform’s efficiency, with the computational process to guide the synthesis of 15 nm TiO 2 NPs taking only 5 min. Feature importance analysis offered mechanistic insights into the nucleation process. Additionally, its versatility was proven by successful extension to ZrO 2 and CeO 2 NPs. In summary, the GBDT-RR framework and its associated synthesis reverse design platform represent a powerful, data-driven tool for the precise and efficient optimization of TiO 2 NP PSDs, facilitating their applications in optical nanocomposites.