While daylight studies vary in scale and complexity, they follow almost identical procedure that yields higher spatial and temporal dimensionality of luminance/illuminance values, which should be reduced into simpler performance metrics. Several complexities arise from depending solely on performance metrics that do not neither agree on unified thresholds, nor report luminous variations spaces might experience due to weather fluctuations. Few research attempts addressed those issues using predefined timesteps. However, they capture instantaneous performance samples of spaces to which they can only be applied, considering fractions of sky conditions that do not offer generalization. This research expands previous endeavors and presents an unprecedented approach that employs unsupervised machine learning to characterize the most representative temporal settings of given locations from widely accessible weather datasets to evaluate internal luminous conditions. The strengths of three algorithms are combined: the ability of Principal Component Analysis to reduce dimensionality, with Self-Organizing Maps and K-means to cluster the reduced data. To exemplify the proposed approach, three locations are investigated, expressing diverse sky conditions. Each of which reflected unique clustering patterns that were translated into temporal profile maps for visualization. This would provide architects and policy makers with methodology to facilitate building performance simulation and design.