Real-time event detection is one of the many applications for wireless sensor networks (WSN). Since the sensor nodes are open to everyone, they are vulnerable to attacks. One of the frequent attacks carried out by malicious attackers that shorten the lifespan of the network and generate congestion is the denial-of-service attack. By modelling an intrusion detection or prevention system they categorise the normal and attack traffic with the present state of the network as good or poor, which is possible to identify DoS attacks by seeing the issue as a classification problem on network state. We propose that Recurrent Neural Networks (RNN) can be used to classify the novel, previously undetected variants of attacks to increase the rate of intrusion detection. Compared to the conventional machine learning classifiers, RNN architectures had a low false positive rate. The main factor is that RNN designs may store data for long-term dependencies through time lags and can correct this with information from subsequent connection sequences. In this paper the simulation was run to evaluate the model's effectiveness, and the findings indicate that the RNN as classifier model outperforms other current machine learning models in terms of improving the rate of detection and prevention by classifying the attacks with less positive rate.