You should consider that available algorithms are based on very different calculations (e.g., complemetarity in seed region, energy, and so on), thus they give some other results and there is no possibility to say "this is the best one". I think that you should review the available literature and then choose the algorithm the best for your research. Moreover, some algorithms like TargetScan allow for adjust the stringency value. The more stringent, the less false positives.