Cortical learning algorithm to predict anomalous web services usage

Abstract

Cyber Security has become a significant concern in corporate web usage, which includes risks of both internal and external computer attacks. The purpose of this study is to introduce and investigate an application of the Cortical Learning Algorithm (CLA) in minimizing potential threats to a web server using machine learning based anomaly detection. The results are compared with 2 major and 3 minor state-of-the-art conventional algorithms. Approximately 38,000 web usage samples were collected over a 6-month time period. This data was organized as an input to CLA through encoders to standardize individual data formats. The output of each encoder is assigned priority weights based on the nature and significance of data. CLA performed exceptionally with an almost 99% rate of anomaly detection, checked manually on unsupervised data. It also helped in processing data as quickly as it arrives in a continuous fashion unlike conventional methods that store and process it offline. This spontaneous and effective approach of CLA proves its potential in enhancing security against computer attacks in the corporate sector. Keywords: Cyber Security, Web Services, Cortical Learning Algorithm, Machine Learning.