Development of A Low-Cost Novel CCT Meter using Machine Learning Regression Techniques to Estimate Real-time CCTs

From improved design optimization, lighting management, advanced adaptive control and energy optimization to Human Centric Lighting (HCL), the application of machine learning in Illumination Engineering offers numerous benefits. Correlated Color Temperature (CCT) is the principal property to be considered to design a lighting scheme human centrically. CCT can be measured precisely using costly devices such as spectroradiometer and colorimeter or less accurately by various color space transformers and predefined mathematical models like Mccamy’s model, which yields less accurate prediction, using measured CIE 1931 RGB values. In order to measure the CCT of a light with high precision and cost-effective fashion, a novel low-cost microcontroller-based device is developed to measure the RGB values. 6 regression techniques are compared using various error term like R-Squared Error (R2), Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) with the help of collected R, G, B data from the developed device. Each regression technique is cross validated by stratified $\mathbf{k}$-fold cross validation to scrutinize each scheme’s attainment for selecting the best model out of them. It is found that Gaussian process regression with Matern kernel $(v=1.5)(G P R m)$ gives the best predictions. Besides, same error terms are also identified for established Mccamy’s model with the help of same data set. It is found that GPRm model is performing better than Mccamy’s model in terms of R2 value, however it is observed that RMSE value is slightly higher for GPRm model.