Application of multivariate adaptive regression splines (MARS) to study the colorization occurring in the process of lactulose production following lactose electro-activation

Lactulose is widely used in the food and pharmaceutical industries and is obtained from the chemical and enzymatic isomerization processes of lactose. Recently, electro-activation, a chemical-free process, has been effectively used for lactulose production at ambient temperature. However, the electro-activated (EA) isomerization process produces colored lactose solutions like the conventional chemical isomerization processes, which becomes a concerning issue. In this context, artificial intelligence-based machine learning tools can be pivotal. The current study observed the color development due to the EA isomerization process and established a model using the multivariate adaptive regression spline (MARS). The color development was expressed in terms of total color difference (ΔE ab ) using the seven input parameters, such as reaction type, reaction time, relaxation time, relaxation temperature, L* , a* , and b* . A total of 127 configurations resulting from the combination of input numbers (1-7) were employed to understand the color development using MARS. The K-fold (K=10) cross-validation technique was applied to select the best-fit combination for each input configuration that underscored the non-linear and complex nature of the color development, offering insights about the input parameters. The reaction type, reaction time, a *, and b* were identified as the key parameters of the color development. The corresponding basis functions of the best-fit configurations reflected the interactions and connected the nonlinearities across the input parameters to understand their impact on the color development using the MARS algorithm.