Reliable protection of microgrids faces significant challenges due to the uncertainty of renewable resources and non-radiative operation. This paper presents a novel adaptive protection scheme based on data mining using the K-Nearest Neighbor (KNN) algorithm, which relies only on local measurements (effective values ​​and symmetric components) and does not require a communication infrastructure. Key innovations of this design include complete independence from communication channels (leading to cost reduction and increased reliability), the ability to simultaneously detect and classify faults using local data, and comprehensive consideration of operational uncertainties (including fault tolerance, different levels of renewable generation, load levels, and different microgrid operating modes) in the analysis and training process. For evaluation, a comprehensive database of different scenarios was created using simulation in DIgSILENT software and the KNN algorithm was implemented in Python. The key results confirm the very high efficiency of the proposed method; so that the overall average accuracy of the KNN algorithm in correctly detecting and classifying faults in all protection devices of the test network was about 95.2%. It is worth noting that the accuracy in many protection points was higher than 98% and in some cases reached 100%. These findings indicate that the proposed KNN-based approach provides an effective, accurate, and reliable solution for adaptive protection of microgrids without the need for communications and considering real-world uncertainties.