This document discusses predicting cyber attacks using machine learning techniques. It proposes building a machine learning model for anomaly detection to identify fraudulent, suspicious or unusual network activities that could indicate cyber attacks. The model would be built using data science methods like variable identification on a past dataset to train and test various machine learning algorithms. Performance metrics of different algorithms would be calculated and compared to find the most accurate model for predicting four types of attacks: DOS, R2L, U2R and Probe attacks. A graphical user interface would then display the prediction results of network attack detection.