Firewall logs can provide valuable information about attempted security breaches and attacks. By classifying these logs, security teams can more easily identify patterns and potential threats, allowing them to take proactive measures to protect their systems. This paper presents a thorough analysis of the various features found in firewall log data. It focuses on selecting the most crucial features using feature selection techniques. The recommended feature sets are then assessed using stacking ensemble XGBoost to demonstrate their suitability. In order to compare the effectiveness of different feature selection techniques, we evaluated at least one member from each technique family using the proposed method. The experimental findings indicate that when Feature Shuffle Random Forest is used in conjunction with the proposed approach, the resulting model is the most accurate, achieving a performance evaluation metric of 99.93\%, signifying exceptional accuracy.