Machine Learning and Continuous Authentication

  • Home
  • /
  • Blog List
  • /
  • Machine Learning and Continuous Authentication
Machine Learning and Continuous Authentication
on 23 Oct 2018 14:52 PM
  • Rang Technologies
  • AI / Machine Learning

Securing the information and data using authentication has evolved immensely over time. From simple passwords to complex patterns and now in today,s digital world, with mechanisms like biometric verification, the data is sure to secure. Having said that, the skills and adaptions of fraudulent minded individuals have also increased multifold. This brings us to the question: What Next? How can the authentication mechanisms be improved?

In today's world, we humans are heavily dependent on computers for communication, banking, security applications, and many other things. This increases the chances of malicious attacks, which calls for high security to protect user secured data and unauthorized access. Currently, we use Pins and Passwords for authentication. For higher security, we adapted biometric verification techniques like:

• Fingerprint Recognition
• Finger Vein Recognition
• Retina and Iris Recognition
• Hand/Palm Recognition
• Voice Recognition
• Signature Recognition
• Face Recognition

Such techniques are certainly not enough to stop all malicious activities. Hence, comes the concept of Machine learning and continuous authentication. Using machine learning techniques to continuously authenticate in the background based on behavioral patterns and other sensory data is one of the primary concepts that has emerged in recent years and is being worked upon by many in the industry. This use of machine learning in continuous authentication is also referred as Behavioral Biometrics.

Every person behaves in a completely individual way. The gait with which someone walks, the fluctuations in vocal tone as they speak, and the cadence with which they type are as unique as fingerprints-but are much harder for malicious actors to capture much less duplicate. Behavioral biometrics uses these patterns to authenticate users and protect data. Behavioral biometric tools run on the actual mobile devices and computer systems connected to an organization's data. Each user with valid access automatically generates a behavioral profile that reflects the distinct ways in which he or she interacts with critical systems-gestures such as keystrokes, screen swipes, and mouse movements. Once a user's profile is learned, their gestures are monitored silently, in real time, to continuously authenticate identity. If behavioral patterns that don't match the profile occur, the system can immediately prompt for other forms of authentication, block access, or lock the device down entirely.

Deep learning algorithms can be used to create a model of user's behavior by analyzing the way the user interacts with a platform, such as login times, IP addresses, devices, or even more detailed actions such as typing, clicking and scrolling habits or the use of keyboard shortcuts. These algorithms will silently monitor future interactions and compare them with the behavioral patterns recorded by it and will flag or block the device for further use based on the security levels defined originally.

Some of such examples include a detection of a sudden increase in the download of documents and user data from a cloud application, abnormal access to the emails, abnormal typing and clicking pattern. In such scenarios, the application will ask the user to provide further proof of access or account ownership such as an OTP sent to a mobile device or email address or plugging in a security USB key.

The first wave of behavioral biometrics solutions to see marketplace adoption was focused on mobile computing. While these applications remain essential, over the next several years behavioral biometrics will become a significant security paradigm in non-mobile computing contexts as well. Adoption of behavioral biometrics technologies will continue to increase across organizations of all types to meet the growing need for highly secure authentication that balances increasingly stringent privacy, security, and user experience quality requirements. By 2020, some 26 billion devices will be connected to the IoT. Ranging from personal and household devices to sensitive military and scientific equipment, these units represent a vast ocean of vulnerability spread throughout human environments-for which device-by-device password authentication is implausible.

There are various security providers like Callsign and UnifyID who have started rolling out Adaptive/Risk-based authentication tools and applications which are efficient enough to recognize any type of malicious attack. It's sure it is not the ultimate authentication concept, but it is definitely a step forward for protecting everyday citizen data.