The level of success of spectral decision in cognitive radio networks depends on how good the prediction model of channel usage is in both licensed and unlicensed bands. In this article three different techniques are explored to predict channel state from the point of view of the use given by the primary user and optimization of the best technique is proposed to achieve a closer prediction to reality from the use of actual data of a WiFi network as validation method. The article concludes that our algorithm is optimal for short history windows.