This work proposes a new SVM based islanding detection technique for the microgrid. Considering the ineffectiveness of Vector Surge (VS) Relays to detect islanding under low mismatch of active power, there is a requirement for a new cost-effective islanding detection technique. In this work, a robust data-based model is designed for efficient islanding detection in case of a low mismatch of active power. Accuracy and algorithmic complexity are the two main issues that are needed to be dealt with when data-based models are considered. The work in this paper achieves higher accuracy for islanding detection using a model of reduced complexity. Firstly, multiple features are extracted from continuous waveforms, and then the new features with reduced dimensionality are created using them. Finally, a Support Vector Machine (SVM) model is trained to perform the islanding detection using these new features. The SVM based islanding detection model thus obtained is tested against the new test cases to analyze the robustness of the model in detecting islanding conditions under different mismatches of active power.