Fraud Prevention with Cloud AI Solutions

The objective of many security attacks and data breaches that corporations suffer from is the violation of sensitive information, such as customers' credit card details. Such attacks are often conducted in stealth mode, and so it is difficult to detect such threats using traditional methods. In addition, the amount of data to be monitored often assumes dimensions that cannot be effectively analyzed with just traditional extract, transform, and load (ETL) procedures that are executed on relational databases, which is why it is important to adopt artificial intelligence (AI) solutions that are scalable. By doing this, companies can take advantage of cloud architectures in order to manage big data and leverage predictive analytics methodology.

Credit card fraud represents an important test for the application of AI solutions in the field...

Introducing fraud detection algorithms

In recent years, we have witnessed an increase in fraudulent activities in the financial sector, and particularly in the area of ​​credit card frauds. This is due to the fact that it is rather easy for cybercriminals to set up credit card fraud, and it has, therefore, become important for financial institutions and organizations to be able to promptly identify fraud attempts.

Furthermore, the activity of fraud detection and prevention in the context of credit card fraud is complicated by the fact that this type of fraud assumes global characteristics; that is, it involves different geographical areas as well as a variety of financial institutions and organizations.

Therefore, it is essential to be able to share the information sources that are available within different organizations around the world.

These sources of information...

Predictive analytics for credit card fraud detection

To adequately address the problem of fraud detection, it is necessary to develop predictive analytics models, that is, mathematical models that can identify trends within the data, using a data-driven approach.

Unlike descriptive analytics (whose paradigm is constituted by business intelligence (BI)), which limits itself to classifying the past data on the basis of measures deriving from the application of descriptive statistics (such as sums, averages, variances, and so on), precisely describe the characteristics of the data being analyzed; instead, by looking at the present and past situation, predictive analytics tries to project itself in order to predict future events with a certain degree of probability. It does this by extrapolating hidden patterns within the analyzed data.

Being data-driven, predictive analytics makes...

Getting to know IBM Watson Cloud solutions

The time has come to get to know one of the most interesting cloud-based solutions available on the market, and will allow us to look at a concrete example of credit card fraud detection in action: we are talking about the IBM Watson Cloud solution, which introduces, among the other things, the innovative concept of cognitive computing.

Through cognitive computing, it is possible to emulate the typically human ability of pattern recognition, which allows adequate contextual awareness to be obtained for decision-making.

IBM Watson can be successfully used in various real scenarios; here are few:

• Augmented reality
• Crime prevention
• Customer support
• Facial recognition
• Fraud prevention
• Healthcare and medical diagnosis
• IoT
• Language translation and natural language processing (NLP)
• Malware detection

Before going into detail about the IBM...

Importing sample data and running Jupyter Notebook in the cloud

Now, let's learn how to use the IBM Watson platform. The first thing we need to do is create an account, if we don't have one already; just connect to the IBM Cloud platform home link provided here athttps://dataplatform.cloud.ibm.com/. You will see the following screen:

IBM Watson home page

To proceed with the registration, select Try it for Free (register) as shown in the preceding screenshot. We will be automatically redirected to the registration form, as shown in the following screenshot:

IBM Watson Registration page

Once registration is complete, we can log in again from the home page:

IBM Watson login form

After logging in, we can create a new project:

IBM Watson start by creating a project screen

We can select the type of project we want to create:

IBM Watson project selection

In our case, we...

Evaluating the quality of our predictions

To correctly evaluate the quality of the predictions that were obtained by our classifiers, we cannot be satisfied with just accuracy_score, but must also use other measures, such as the F1 score and the ROC curve, which we previously encountered in Chapter 5, Network Anomalies Detection with AI, dealing with the topic related to anomaly detection.

F1 value

For the convenience, let's briefly go over the metrics that were previously introduced and their definitions:

Sensitivity or True Positive Rate (TPR) = True Positive / (True Positive + False Negative);

Here, sensitivity is also known as the recall rate:

False Positive Rate (FPR) = False Positive / (False Positive + True Negative...

Summary

We have learned how to develop a predictive model for credit card fraud detection, exploiting the IBM Cloud platform with IBM Watson Studio.

By leveraging the IBM Cloud platform, we have also learned how to address the issues related to the presence of unbalanced and non-stationary data within the dataset concerning credit card transactions and made full use of ensemble learning and data sampling techniques.

In the next chapter, we will delve deep into generative adversarial networks (GANs).