Bootstrapping Adaptive Linear Neuron in Near Infrared Spectroscopic Analysis

Abstract

Near infrared spectroscopic analysis requires a predictive model to extract relevant information from a complex near infrared spectral data so that the internal composition of products can be measured indirectly. Even though ensemble models show a better predictive performance compared to that of a single model in most cases, the computational cost will be multiplied for building multiple models. Besides, a combination of several different sub-models causes an ensemble model to be much complex than a single model. Thus, this study proposes the bootstrapping adaptive linear neuron (Adaline) that adapts the philosophy of bootstrapping aggregation approach. Without changing the architecture of an Adaline, the results indicate that the proposed the bootstrapping Adaline is promising to achieve a better performance than an Adaline with an average 18.6% improvement. This suggests that the bootstrapping algorithm is promising to enhance the predictive accuracy of the Adaline model in near infrared spectroscopic analysis.