Machine Learning-Guided Design of MQ Silicone Resin Reinforced Addition-Curing Silicone Rubber: From Literature Data Mining to Experimental Validation

MQ silicone resins are widely used reinforcing fillers for addition-curing liquid silicone rubber (LSR); however, establishing a quantitative composition–property relationship remains challenging because published data are fragmented across matrix chemistries, crosslinkers, catalyst systems and testing standards. Here we present a machine-learning-guided workflow integrating literature data mining, interpretable random-forest (RF) modelling and independent experimental validation for the design of MQ-reinforced LSR. An RF model trained on 55 curated literature points spanning RTV and LSR systems, using four physically motivated descriptors (MQ content, M/Q ratio, curing system and vinyl content), yielded leave-one-out coefficient of determination (R2) values of 0.741 for tensile strength (TS) and 0.730 for Shore A hardness (HA), with mean absolute errors of 0.63 MPa and 8.95 ShA, respectively. Feature-importance and partial-dependence analyses identified MQ content as the dominant descriptor. Guided by the model, seven LSR formulations (vinyl content 4 wt%, M/Q = 0.8, loading 5–35 wt%) were designed and fully characterised: the model reproduced the measured TS and HA for all seven formulations within the corresponding training mean-absolute-error tolerance, whereas elongation at break (EB), whose prediction is substantially weaker (LOO R2 ≈ 0), was captured only as a qualitative trend with respect to MQ loading. This workflow demonstrates that a modest, curated literature dataset, mined by an interpretable ML model, can support formulation design and independent experimental validation—an efficient, low-cost alternative to trial-and-error optimisation.