Unsupervised learning methods are not able to deal with such scenario unless they are converted to supervised ones. If so, then use classical classification process.
Samer Sarsam thanks for reply, if we cluster the dev data without their target values further check each dev data value including their target and with their corresponding clusters,
if it is possible to find which one is in wrong cluster??
If you have instances with labels (classes), you can cluster them first. After that, use the classes to cluster evaluations. Meaning that, compare isntances in each cluster and their original target labels. Instances that originally belong to class A (for instance) but they don't belong to cluster A are considered to be the possible errors.
However, if you have the dataset with labels, then directly classify it (no need to cluster it). Misclassified instances are considered to the classification errors.
Samer Sarsam, I only have the training set and exactly, my problem is to find the data entries in it which have wrong labels or target values, then in this case I would have to proceed with this clustering method only, right?? so that the classification error can be minimized.