DDoS attacks are one of the most serious threats against network security and availability. Sockstress attack is one of them which could simply exhaust the server resources by opening many established yet useless TCP connections that advertize a zero window size, the server being obliged to keep these idle sessions open indefinitely. This paper presents a Neural Network based framework for detecting Sockstress DDoS attacks. The UNSW-NB15 dataset is selected for training and testing the system with concerns for robust feature engineering for the related Sockstress attacks. Results show that the proposed system is effective, in terms of accuracy, precision, recall, and F1-score, for detecting Sockstress attacks.