Modeling and analysis of cooperative spectrum sensing is an important aspect in cognitive radio systems. In this paper, the problem of energy detection (ED) of an unknown signal over Nakagami-$m$ fading is revisited. Specifically, an analytical expression for the local probability of detection is derived, while using the approach of ED at the individual secondary user (SU), a new fusion rule, based on the likelihood ratio test, is presented. The channels between the primary user to SUs and SUs to fusion center are considered to be independent Nakagami-$m$. The proposed fusion rule uses the channel statistics, instead of the instantaneous channel state information, and is based on the Neyman-Pearson criteria. Closed-form solutions for the system-level probability of detection and probability of false alarm are also derived. Furthermore, a closed-form expression for the optimal number of cooperative SUs, needed to minimize the total error rate, is presented. The usefulness of factor graph and sum-product-algorithm models for computing likelihoods, is also discussed to highlight its advantage, in terms of computational cost. The performance of the proposed schemes have been evaluated both by analysis and simulations. Results show that the proposed rules perform well over a wide range of the signal-to-noise ratio.