They have different targets because they have different sequences, they are the two sides of the pre-miRNA stem-loop. Try to see in your data which side is expressed (if you have the sequences of mature miRNA) or literature.

How you get the differentially expressed miRNAs? If it is by sequencing you should have to have the sequence. Then, you can look for it in miRBAse and check which arm is coming from. I would strongly recommend to do not base your target predictions in only one method. Use at least two of them and consider only those targets that come up using both methods. This is because the high positive false results that you obtain from the target prediction algorithms.