Cyber security of internet-based cloud networks is a major concern for companies. Networks are always open to intrusions from both inside and outside of organizations. However, using traditional techniques to identify behavioral changes or malicious attacks is challenging. In this paper, a novel DEEP learning-based FIREwall tuning (Deep Fire) method has been developed to integrate deep learning and honeypot in order to mitigate and prevent the attacks of cloud systems. In order to identify ransomware activity and attack patterns, Apache spark engine is used which combines information from TrAck Replay Evaluate (TAPE) systems. The proposed technique uses Convolutional neural Network (CNN) for detecting the intrusion into two classes such as attack and normal. The proposed system has been evaluated using python simulator. The proposed technique has been evaluated using specific parameters such as detection accuracy and false alarm rate. The proposed system achieves higher detection accuracy of 22.6%,16.6% and 2.38% than the existing systems such as IDPS, virtual machine (ICI) and Honeypot based IDS technique respectively. By utilizing Apache Spark, the proposed DEEP-FIRE correlates network features, host attributes, and various events from other systems like TAPE and firewall to produce more accurate results.