Malware detection has gained huge attention in recent times. This is mainly because of the increase in new malware variants which pose a significant threat to information security. The conventional malware detection systems are not capable of detecting new generation malwares due to the constant changes in the network behavior. An efficient malware detection approach must be able to handle the dynamic changes in the malware behavior with a very minimum processing time to identify malicious attacks at the initial stage. This paper presents a novel Performance Importance Weighted Random Forest (PERI-WRF) for detecting different types of malwares in network systems. The proposed PERI-WRF incorporates a novel data reduction technique which is capable of reducing the size of training data to maximize the classification accuracy. A clustering algorithm consisting of GWO and K-means++ algorithm is implemented to group the malicious data samples collected from input. To validate the effectiveness of the detection framework, the system was tested using various evaluation metrics. Results show that the proposed malware detection model with novel data reduction techniques achieves superior classification accuracy, and the proposed approach is appropriate for detecting real-time malwares with superior accuracy and low MAE score.