Fast and accurate detection and identification of power line outage is of paramount importance for the prevention of cascading failures in power systems, as well as prompt and effective restoration following the outage. Traditional approaches can only detect single and double line outages due to the combinatorial complexity challenges involved in the algorithms. A novel approach is to cast it to a sparse vector estimation problem, which can be solved efficiently by taking advantage of the recent progress in compressive sensing and variable selection. In this work, we adopt a similar approach to formulate the problem as a sparse binary-valued vector estimation problem, and leverage the cluster structure existing in most multiple-line outages to solve it. We propose two low-complexity graph-based algorithms to identify clustered line-outages. Simulated tests in IEEE-118 bus system confirm that the proposed algorithms can significantly improve the accuracy and efficiency of baseline algorithms that do not leverage the cluster structure of multiple-line outages.