A Hybrid Machine Learning Approach to Shelflife Prediction of <scp>NBR</scp> for Long‐Term Storage Applications

ABSTRACT This study examines the thermal degradation kinetics and shelf‐life prediction of Nitrile Butadiene Rubber (NBR) using thermogravimetric analysis and a hybrid machine learning framework. Both virgin and two‐year‐aged NBR samples were assessed with model‐free kinetic methods: Ozawa–Flynn–Wall (OFW), Kissinger–Akahira–Sunose, and Kissinger. The activation energy for virgin samples ranged from 67.26 to 73.33 kJ/mol; for aged samples, it was slightly lower (66.9–72.57 kJ/mol), indicating molecular degradation. Shelf life estimated at 40°C with Toop's equation decreased from 111.29 to 89.01 years under OFW kinetics. An Arrhenius‐type exponential degradation model, using a Random Forest (RF)–AdaBoost algorithm (80/20 train‐test split), achieved high accuracy ( R 2 &gt; 0.999, RMSE &lt; 1%, MAPE &lt; 2%). The fitted parameters (Ψ ≈0.043°C −1 ; Υ = 0.08–0.11; Φ ≈1) and deviations of less than 1% from the experimental shelf life confirm the robustness of the nonlinear degradation prediction.