In most cases, continuous authentication avoids using passwords, access patterns, biometric recognition, etc. when the user wish to have access to an app or service requiring authentication. In this sense, permanent authentication increases users’ security regarding the operations executed on the device. Moreover, we can take advantage of this continuous trust status to make user app interactions much simpler and more fluent by doing so, users’ experience gets better.
Despite the advantages of these continuous-authentication systems, current solutions raise a number of challenges, for instance: selecting the dimensions and features that allow to shape the owner’s behavior and be able to clearly and precisely discern its behavior from other users’ one; enabling system adaptability to slight changes in user’s behavior; reducing authentication time; using new functionalities or optimizing device resources’ use and consumption. These aspects are critical to provide the user with a satisfactory experience and not excessively impact the battery.
The system functioning is divided into four phases:
- Phase 0. Over this phase the most relevant dimensions and features intended to shape user’s behavior are selected. It should be highlighted that this selection process is a one-time process performed prior to system development over the design phase.
- Phase 1. Acquisition of the mobile device data to extract the predetermined features and create a dataset where such features will be stored. Data collection is periodically performed in one-minute cycles for two weeks.
- Phase 2. Firstly, Machine Learning algorithm is trained by means of the generated dataset in order to shape a profile for user’s behavior. Once this training has finished, the evaluation phase is triggered, over which the system compares the current user’s behavior with the one stored over the training phase. By doing so, the system returns an authentication level ranging from 0.0 and 0.1. This is a one-minute process as well.
- Phase 3. System adaptability to new changes in user’s behavior by insertion and removal of vectors within the dataset, keeping it updated and preventing system from overtraining.
These phases, together with some steps in detail, are shown in the following figure.
In the following video you will find further details on the design and functioning of this system: