Network　security　has　emerged　as　a　crucial　universal　issue　that　affects　enterprises,　governments,　and　individuals.　The　strategies　utilized　by　the　attackers　are　continuing　to　evolve,　and　therefore　the　rate　of　attacks　targeting　the　network　system　has　expanded　dramatically.　An　Intrusion　Detection　System　(IDS)　is　one　of　the　significant　defense　solutions　against　sophisticated　cyberattacks.　However,　the　challenge　of　improving　the　accuracy,　detection　rate,　and　minimal　false　alarms　of　the　IDS　continues.　This　paper　proposes　a　robust　and　effective　intrusion　detection　framework　based　on　the　ensemble　learning　technique　using　eXtreme　Gradient　Boosting　(XGBoost)　and　an　embedded　feature　selection　method.　Further,　the　best　uniform　feature　subset　is　extracted　using　the　up-to-date　real-world　intrusion　dataset　Canadian　Institute　for　Cybersecurity　Intrusion　Detection　(CICIDS2017)　for　all　attacks.　The　proposed　IDS　framework　has　successfully　exceeded　several　evaluations　on　a　big　test　dataset　over　both　multi　and　binary　classification.　The　achieved　results　are　promising　on　various　measurements　with　an　accuracy　overall,　precision,　detection　rate,　specificity,　F-score,　false-negative　rate,　false-positive　rate,　error　rate,　and　The　Area　Under　the　Curve　(AUC)　scores　of　99.86%,　99.69%,　99.75%,　99.69%,　99.72%,　0.17%,　0.2%,　0.14%,　and　99.72　respectively　for　abnormal　class.　Moreover,　the　achieved　results　of　multi-classification　are　also　remarkable　and　impressively　great　on　all　performance　metrics.