Integrating inverter-based renewables in electrical power systems introduces several challenges to different power subsystems. One of these is the protection system. The conventional protection system functions may not properly function for different fault scenarios due to the inverters’ behaviors. The protection system elements of the transmission lines connected to inverter-based generators need to be designed to cope with the different changes experienced by the power system, such as the penetration level, the reduction of inertia, the lack of zero and negative-sequence components, and the changing source impedance. Different techniques can be used to mitigate these changes. Adaptive protection schemes are the most common of these. The proposed adaptive schemes in this research are Machine Learning (ML)-based. They are mostly related to detecting and classifying electrical faults in the transmission lines connecting inverter-based generators, namely photovoltaic (PV) systems and doublyfed induction generator (DFIG) wind turbines. The 39-bus power system was used as the transmission system network and to extract the power signals. Two adaptive methodologies were considered. The first selects the appropriate pretrained and pre-optimized classifier by monitoring the circuit breaker status and the active power of the generators' outputs behind the protected transmission line to identify the power system topology. This is basically a lookup table approach that inputs the circuit breaker's status (to identify which generator is connected to the grid) and the output active power of the generator (to detect which generators are active). The second adaptation scheme uses incremental learning as an adaptation mechanism. The incremental learning process uses the measured voltages and currents online (data stream) to update the parameters of the classification model associated with data drift detection and self-labeling models. This data-driven approach is used to minimize the time in the updating mode, which reduces computational complexity and increases the speed of detection and classification. The main aim of the incremental is to maintain the efficiency of the fault detector in the face of changing the statistical characteristics of the data and to do that without human interference. The main features of this research can be summarized as follows. It updates the literature on the impact of inverter-based generators on power system protection elements and the mitigation approaches concentrating on using artificial intelligence-based techniques. In addition, it utilizes a wide range of ML algorithms to detect and classify different types of faults in transmission lines connected to IBGs. Furthermore, it investigates the impact of data imbalance, which is essential when considering power system faults as minority events. It also proposes two fault detection and classification techniques for adaptive protection design approaches considering the ML algorithms. Finally, it deals with data streams by having data drift detection and labeling using Hoeffding’s inequality theory and self-training self-labeling semi-supervised model as part of the ML-based power system protection scheme. When compared to similar methodologies from the literature, the findings demonstrated a competitive performance. The incremental learning strategy with the suggested ftting and updating criteria has exceptional performance in terms of retraining time and detection and classification accuracy.