This study proposes a sensor-driven UV-C disinfection system enhanced by intelligent optimization algorithms for application in library environments. The architecture combines real-time multi-sensor monitoring with a hybrid machine learning framework integrating Random Forest regression and Response Surface Methodology. Based on 42 experimental runs, a predictive model was trained to estimate microbial log-reduction and material impact as a function of operational parameters. The system implements closed-loop adaptive control to dynamically regulate irradiance, exposure time, and airflow. Experimental validation using E. coli, S. aureus, A. niger, and MS2 demonstrated 5.2-5.8 log10 reductions within 30-45 seconds. The predictive model achieved R² > 0.90 for log-reduction estimation, enabling reliable parameter optimization under preservation constraints (ΔE < 1.5; tensile loss < 2.5%). Compared to a conventional fixed-dose UV baseline configuration, the proposed intelligent architecture reduced processing time by 40% and energy consumption by 35%. The results highlight the potential of data-driven adaptive control in transforming UV-C disinfection from a static process into a scalable applied IT solution for preventive conservation.