Machine Learning-Based Recommender System for Pulsed Laser Ablation in Liquid: Recommendation of Optimal Processing Parameters for Targeted Nanoparticle Size and Concentration Using Cosine Similarity and KNN Models

Achieving targeted nanoparticle (NP) size and concentration combinations in Pulsed Laser Ablation in Liquid (PLAL) remains a challenge due to the highly nonlinear relationships between laser processing parameters and NP properties. Despite the promise of PLAL as a surfactant-free, scalable synthesis method, its industrial adoption is hindered by empirical trial-and-error approaches and the lack of predictive tools. The current literature offers limited application of machine learning (ML), particularly recommender systems, in PLAL optimization and automation. This study addresses this gap by introducing a ML-based recommender system trained on a 3 × 3 design of experiments with three replicates covering variables, such as fluence (1.83–1.91 J/cm2), ablation time (5–25 min), and laser scan speed (3000–3500 mm/s), in producing magnesium nanoparticles from powders. Multiple ML models were evaluated, including K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), Random Forest, and Decision trees. The DT model achieved the best performance for predicting the NP size with a mean percentage error (MPE) of 10%. The XGBoost model was optimal for predicting the NP concentration attaining a competitive MPE of 2%. KNN and Cosine similarity recommender systems were developed based on a database generated by the ML predictions. This intelligent, data-driven framework demonstrates the potential of ML-guided PLAL for scalable, precise NP fabrication in industrial applications.