This study investigates the performance of an infrared drying system using fixed (FTM) and two-step temperature methods (TTM) in preserving the phytochemical and color attributes of spearmint leaves, alongside an energy analysis and the development of a machine learning (ML)-based predictive model. Results showed that color degradation as a quality parameter intensified with rising temperature, peaking at 72 ◦C (ΔE = 16.68) under FTM, whereas TTM, which applies a strategic 10 ◦C temperature drop upon reaching the falling rate period, significantly mitigated discoloration, e.g., reducing ΔE from 16.36 (70 ◦C) to 13.69 (70to60◦ C). GC analysis revealed that 48to38◦ C of TTM yielded the highest concentrations of key phytochemical constituents, while 68 ◦C of FTM demonstrated superior retention among fixed-temperature treatments. Principal component analysis (PCA) discriminated effectively between drying protocols, with the first two components explaining 72 % of the variance and highlighting distinct phytochemical and color degradation patterns. TTM consumes 4.98–8.45 % less energy than the corresponding fixed process at the second-step temperature (58to48◦C vs. 48 ◦C; 72to62◦ C vs. 62 ◦C, respectively) and reduces color degradation by 9.5–16.5 % compared to the fixed process at the first-step temperature (72to62◦ C vs. 72 ◦C and 58to48◦C vs. 58 ◦C), establishing a dual-reference benchmark. Three ML algorithms, Least Squares Boosting (LSBoost), Random Forest (RF), and Support Vector Machine (SVM), were employed to develop predictive models for key phytochemicals. LSBoost provided the highest predictive accuracy for 1,8-Cineole and Pulegone, achieving R2 values of 97 % and 96 %, respectively, whereas RF yielded the most accurate predictions for Menthone with an R2 of 98 %. In terms of computational performance, RF exhibited the lowest CPU time and highest overall efficiency, followed by LSBoost and SVM. Bayesian-optimized hyperparameters enhanced model generalization, validated via k-fold cross-validation and independent testing (R2: 0.88–0.90, RMSE: 0.39–1.84).
