This paper presents a novel deep learning framework that integrates experimental measurements with advanced modeling techniques to predict key optical parameters, including luminous flux, correlated color temperature (CCT), and chromaticity coordinates of white-red light-emitting diodes (LED) configurations under diverse operating conditions. The heatsink temperature, white LED driving current, and red LED driving current were each varied systematically to generate a comprehensive set of 5,166 spectral power distribution (SPD) measurements. This dataset, partitioned into training (4,182 data sets) and testing (984 data sets) sets, encapsulates the complex physical mechanisms influencing LED performance, such as temperature-induced spectral shifts and current-dependent optical behavior. Four deep learning algorithms were evaluated. Each model was trained to reconstruct the SPD curves and predict the corresponding optical and chromatic parameters. Our results indicate that Long Short-Term Memory Network (LSTM), Convolutional Neural Network (CNN), and Autoencoder (AE) outperform Backpropagation Neural Network (BP-NN), with CNN achieving the highest accuracy in predicting SPD curves and LSTM achieving the highest accuracy in predicting the optical and chromatic parameters. Furthermore, By mimicking the effects of varying red phosphor ratios through independent control of red LED output, our approach can provide deeper insights into the underlying physical phenomena governing LED spectral behavior. This integrated methodology not only enhances our understanding of the interplay between operating conditions and LED performance but also offers a robust predictive tool for the design and optimization of next-generation LED lighting technologies.