In this paper, the performance analysis and concepts of the area under the ROC curve (AUC) in energy detectionbased cognitive radio system are presented and investigated. The small-scale variations of the fading signal under both lightof-sight (LOS) and multipath scenarios, modeled by the κ-µ distribution, are considered with noise uncertainty. Moreover, admitting the presence of noise power estimation error as well as the usage of diversity schemes, the AUC are also outspread and studied. Finally, comparisons are performed and show an excellent improvement in the overall detection capability of the cognitive radio system, considering the effects of noise uncertainty when the most common diversity schemes are implemented.