In cognitive radio systems, the accuracy of spectrum sensing depends on the received primary signal strength at the secondary user (SU). In fact, a single node sensing would be compromised if the received signal strength is not high enough to be detected by this node. In this paper, we propose a cooperative decision fusion rule based on adaptive linear combiner. The weights which correspond to confidence levels affected to SUs, are determined adaptively using the Normalized Least Mean Squares (NLMS) and the Recursive Mean Squares (RLS) algorithms. The proposed algorithms combine the SUs decisions with the adaptive confidence levels to track the surrounding environment. Simulation results show a high adaptability of the proposed scheme, as the operating conditions change. Furthermore, the proposed algorithms do not necessitate a prior knowledge about the PU features and are very efficient compared to conventional decision fusion techniques.