<p>Microgrids (MGs) are localized energy systems that encourage the use of renewable sources, which may be operated with support by the main grid. MGs fault detection is challenging due to its bidirectional current flow with varying fault currents. In this research work, the authors proposed an advanced fault detection and classification method using a fully connected neural network, “Multilayer perceptron” (MLP). This model is a supervised generative model. The suggested model is a subset of labeled dataset that is partitioned into two sets: train sets and test sets. The proposed architecture includes an Input layer (3 neurons), Hidden layer (100 neurons), Rectified Linear Unit Activation Function Layer, Hidden layer (50 neurons), Rectified Linear Unit Activation Function Layer, Output Layer, and Sigmoid Function. The proposed model is verified in an IEEE-modified 7 Bus MG system designed in RSCAD by creating faults at various locations of the MG. RSCAD offers a highly realistic environment for implementing practical relays within a MG environment. The proposed technique can detect and classify both symmetrical and unsymmetrical faults accurately. The proposed technique’s competence is compared with other techniques like Naïve Bayes (NB), Decision Tree (DT) and SVM. The suggested MLP shows an accuracy of 97%. The suggested deep learning-based architecture can automatically understand hierarchical data, which eliminates feature engineering in dealing with large datasets like fault current and voltages (datasets available in a relay) in MG. Once the model is trained, it can detect the fault accurately in any MG structure and protect the MG from blackouts or unwanted tripping.</p>