Machine Learning May Be the Solution to Enterprise Security Woes
By Karl Zimmerman

For large enterprise organizations, it can be next-to-impossible to identify attacks and act to mitigate them in good time. That's one of the reasons executives often discover security breaches when an external researcher - or worse, a journalist - gets in touch to ask why hundreds of millions of logins for their company's services are freely available on hacker forums.

The huge volume of incoming connections, the heterogeneity of services, and the desire to avoid false positives leave enterprise security teams in a difficult spot. Finding potential security breaches is like finding a tiny needle in a very large haystack - monitoring millions of connections over thousands of servers is not something that can be managed by a team of humans.

Enterprise security is often preventative: we build a system that - we hope - reduces security risks as much as possible and deploy simple pattern matching intrusion detection systems, crossing our fingers and hoping nothing gets through.

It's not that we lack data about attacks; if fact, we have too much of it. What we lack is an intelligent system that can analyze huge volumes of data and extract actionable intelligence about security threats without a an overwhelming proportion of false positives. If the signal-to-noise ratio is too low, all we've done is to replace a huge haystack with a slightly smaller one.

One possible solution, as you might have guessed, is machine learning. Machine learning algorithms, trained on the characteristics of particular networks, are likely to be far more successful at identifying real threats and notifying the right people.

That's the basic idea behind tools like Apache Spot, an advanced threat detection system that uses machine learning to "analyze billions of events in order to detect unknown threats, insider threats, and gain a new level of visibility into the network."

Spot - which runs on top of Hadoop - uses a variety of techniques, including machine learning, whitelisting, and noise filtering to monitor data from network traffic, filter bad traffic from good, and generate a shortlist of potential security threats.

Spot uses an open data model for threats, making it relatively easy to integrate the data it produces with existing tools and to collaborate with other organizations.

Apache Spot was recently open sourced by Intel and Cloudera, and accepted as an Apache project. It was originally an Intel project called Open Network Insight (ONI). A number of other large organizations have been contributing to Spot since it was open sourced. The hope is that an open source project using a common data model will gain traction in enterprise organizations, who can collaborate to help reduce the devastating, and expensive, impact of security breaches.