The present research paper is concerned with the application of the modern machine learning algorithms for defending electrical power system faults, which is an important field in order to maintain the proper functionality of the system. Time-domain and frequency-based fault detection techniques while efficient are not fully capable of addressing the growing challenges in modern power systems. Basing on the subject of fault detection, it is focused on investigating with the help of several machine learning models such as SVM, Random forest, ANN, and XG boost techniques how their implementation can enhance the accuracy and time frame. The method used in the research entailed data gathering and accumulation from different sources thereafter proceeding to normalize and extract features from such data. Cross validation methods were used in training and assessing the models with performance measures such as accuracy, precision, recall and F1 score. The outcomes revealed that the performance of XGBoost is better than other models in terms of fault detection accuracy and fault detection reliability. The confusion matrix also supported the results and the ROC-AUC analysis provided the exact confirmation regarding the power of XGBoost in minimizing the false positives and in maximizing the true positives. However, the complicated computational issue and accurate data-set are the critical issues even in the current practice. Based on this paper, machines learning based approach for fault detection is more effective as compared to the conventional method especially the XGBoost ensemble method. Thus, the study also calls for additional research that will improve these models for real-time applications in order to address existing issues in relation to the expansion and development of more robust and flexible power systems in the future. This work lays the groundwork for introducing machine learning into the power system for the purpose of improving on fault diagnosis and the stability of the system.