The paper proposes a new machine learning approach for cyber security in big data. It combines multiple classifiers into an ensemble "outfit" approach. The outfit approach achieves 99.8% accuracy in distinguishing benign from malicious web pages, outperforming individual classifiers. The methodology collects and prepares a big dataset to train and evaluate KNN, SVM, MLP classifiers. Results show the outfit approach has higher true positive rate, F-measure, and recall while lowering the false positive rate compared to individual classifiers. The research aims to better detect cyber threats and improve security of big data.