Machine learning-driven enhancement of economical color sensors for accurate light source categorization

With the popularity of IoT devices, the built-in color sensor offers potential expansion possibilities. This study aims to enhance the economical color sensors for light source classification through machine learning. Employing a broad collection and analysis of data from 142 data points across 11 distinct light sources, we compared five machine learning classification models, including decision tree (DT), support vector machine (SVM), linear discriminant analysis (LDA), k-nearest neighbor (KNN), and naive bayes (NB) models. The finding revealed that DT exhibited superior performance (accuracy = 96.48%), followed by NB, LDA, and SVM, while KNN was relatively less effective. Moreover, DT was demonstrated to be the most efficient in visual categorization, especially suited for classifying light sources based on the values of R, G, B, and C channels. To further verify the applicability of the proposed DT classification rules, mixed light source validation experiments were conducted by randomly combining data from two and three types of light sources. The results demonstrated that the DT model maintained 92% accuracy for two-type mixed sources and 82.98% accuracy for three-type mixed sources, confirming its robustness even under complex lighting conditions. This study provides a solid theoretical foundation and practical guidelines for future challenges in addressing light source metamerism and spectral fitting.