The effect of misclassification of authentic profiles is assessed by examining the
MAE of the system before and after detection and filtering. The increase in MAE is
observed to be less than 0.05 on a rating scale of 1–5. Finally the effectiveness of
the attack as measured by the prediction shift on the targeted item is shown to be
significantly reduced when detection is used. All three classifiers reduce the range of
attacks that are successful, particularly at low attack sizes. The SVM algorithm, in
particular, dominates for attack sizes less than 10%, allowing no resulting prediction
shift over that entire range.