Over the last few years, cloud computing has been the buzz. Cloud computing services offer an infrastructure that is highly scalable and supports high-performance computing. With high adoption by businesses of all sizes. Development and deployment of applications within the cloud platform are easy and time to market is done in a fraction of the time.

Artificial intelligence is not a new technology. It has been here for a long time and has helped develop computers and software that perform tasks that are associated with intelligence. Machine learning and deep learning are subsets of artificial intelligence that involve the development of algorithms that learn from data inputs and give intelligent output based on that data and the learned patterns.

A lot of research has been done and still is being done on implementing artificial intelligence into cloud computing. Cloud service providers such as Amazon, Google and Microsoft have already integrated AI into their clouds to improve service delivery. AI brings about capabilities such as machine learning, recognition of patterns and robotics to the cloud. On the other hand, the cloud is able to provide a wide range and large volumes of data since these capabilities are largely dependent on data as input so as to produce the desired output. The cloud also allows the systems to open-access and open-source data which is very crucial in facilitating collaborative learning.

One area where AI can be implemented is real-time fraud detection in the cloud for online banking transactions. The cloud environment can be used in deploying the learning algorithms that help in detecting fraudulent transactions in the online banking system and report them in real-time. In traditional systems, fraud detection depended on analysis of data where algorithms would detect transactions that seemed suspicions and the card holders would be called to ascertain whether they made the transaction. This was reactive and not pro-active and led to a loss of billions of dollars.  Since there is an increase in online banking transactions, such systems are often very complex with a myriad of different solutions blended together, thus the need for better technologies that can detect fraud as it happens and stop it before it is successful.

Machine learning and deep learning have provided the techniques to detect fraud within the cloud environment giving rise to intelligent clouds. Algorithms to detect fraud have been used but they need a real-time agent to be able to perform in real-time. An agent, in AI, in an autonomous entity that receives input and gives output after it has executed a set of instructions. To obtain a real-time system, several agents have to be connected to form an agency. The algorithms are fed with a lot of data on normal transactions by the card owner. They are able to learn the transaction sequences, amounts, types all the patterns that arise. This helps in detecting transactions that do not conform to the patterns.

The input data used in fraud detection systems card holder’s financial and demographic information such as income levels, address (both residential and office), bank transaction history such as shopping and bankruptcy status among others. This forms the analytical data set to be used by the algorithms. The data sets are analyzed using several techniques including self-organizing maps, logistic regression and support vector machines. The logistic regression uses a probabilistic score for each transaction to ascertain the likelihood of being illegitimate.  The algorithms run on a self-organizing machine that learns without supervision (unsupervised) by searching for the patterns that appear from series of transactions. The support virtual machine implements supervised learning algorithms that classify transactions as fraudulent or legitimate. When used together, these three mechanisms ensure that the algorithms learn the past and analyze the present in real-time and what is learned is sent to the fraud detection mechanism making it more informed and accurate with time.

These learning algorithms are executed within a short time and the results produced in a fraction of a second. The system then is able to receive transaction data as inputs, detect and classify the transactions as legitimate or fraudulent and produce output in real-time.  If a transaction is found to be fraudulent, the system is able to block and report it to the merchant bank and card holder even before it is successful.

Improvement of the process of detecting fraudulent transactions can be improved by running several algorithms in parallel, collecting their outputs and aggregating them in real-time. When the real-time applications are deployed within the cloud, it makes them highly scalable with proper load balancing and maintenance of the hardware. Since a real-time system is resource intensive, the cloud environment is very suitable so that they can be available on demand, thus the time for data acquisition and analyzation is greatly reduced.  The cloud makes advanced solutions readily available to anyone allowing you to easily deploy complex AI and machine learning solutions.

Let me know if you have any ideas, or have started your cloud journey or if you have any input or just want to talk you can reach me on twitter @UlvBjornsson or hit me up here on the blog!

See you around,

Ulv