With the tremendous advances of the wireless devices technology, securing wireless sensor networks became more and more a vital but also a challenging task. In this paper we propose an integrated strategy that is meant to discover malicious nodes within a sensor network and to expel them from the network using a node self-destruction procedure. Basically, we will compare every sensor reading with its estimated values provided by two predictors: an autoregressive predictor [1] that uses past values provided by the sensor under investigation and a neural predictor that uses past values provided by adjacent nodes. In case the absolute difference between the measured and the estimated values are greater then a chosen threshold, the sensor node becomes suspicious and a decision block is activated. If this block decides that the node is malicious, a self-destruction procedure will be started.