Predictive and Interpretive Modeling of Dual‑Function LPG System Performance with Ridge Regression and Random Forest

Dual function liquefied petroleum gas (LPG) systems that provide simultaneous cooking and refrigeration represent an emerging energy technology, yet the complex interactions governing their performance remain unquantified. This study applies Ridge Regression and Random Forest analysis to an experimental dataset of 4,751 observations, with predictors standardized and multicollinearity assessed via variance inflation factor, to model the effects of operational runtime (RT), mass flow rate (MFR), pressure drop (ΔP), and evaporator load condition (LC) on system outputs. The Ridge regression models for coefficient of performance (COP), refrigerating effect (RE), and evaporator chamber temperature (ECT) achieved strong predictive accuracy. The flame temperature (FT) model, however, showed limited predictive performance and is therefore treated as an interpretive tool to reveal directional effects and relative parameter influences rather than for precise point predictions. The analysis identified MFR as the primary performance driver, substantially enhancing both RE and FT, while ΔP severely reduced FT and provided a minimal refrigeration benefit. Prolonged RT gradually improved COP but decreased RE and FT, consistent with fuel depletion. Operating under a water load increased all outputs at the expense of higher ECT. Random Forest analysis, supported by cross validated permutation importance, confirmed MFR as the dominant factor for RE and ΔP as the most influential for FT. The findings demonstrate that system behavior is optimized by maximizing MFR while minimizing ΔP, and that loaded operation improves performance with an accompanying rise in ECT. The resulting framework provides both predictive equations for key outputs and interpretive insights for flame temperature, offering a foundation for understanding, optimizing, and designing efficient integrated thermal systems.