Fraud is a multibillion dollar business and it is increasing every year. PwC's global economic crime survey of 2016 shows that more than one in three (36%) organizations are experiencing economic crime.
The possibility of fraud evolved with technology, in particular. Information technology Business reengineering, reorganization or downsizing can weaken or eliminate control, while new information systems can present additional opportunities for fraud.
The traditional method of data analysis has long been used to detect cheating . They require a complex and time-consuming investigation that deals with different knowledge domains such as finance, economics, business practice and law. Fraud often consists of many incidents or incidents involving repeated violations using the same method. Examples of fraud can be similar in content and appearance but are usually not identical.
The first industries to use data analysis techniques to prevent fraud were telephone companies, insurance companies and banks (Decker 1998). One of the earliest examples of successful application of data analysis techniques in the banking industry is the FICO Falcon scam assessment system, which is based on a neural network shell.
The retail industry is also experiencing fraud in POS. Some supermarkets are beginning to use digital closed-circuit television (CCTV) along with POS data from transactions that are most vulnerable to fraud.
Recent Internet transactions are causing great concern, with some research showing that internet transaction scams are 12 times higher than in-store fraud.
Fraud involving cell phones, insurance claims, tax refund claims, credit card transactions, etc. Is an important issue for governments and businesses, but detecting and preventing fraud is no easy task. Fraud is an adaptive crime, so a special method of intelligent data analysis is needed to detect and prevent it. This method is in the Knowledge in Database (KDD), Data Mining, Machine Learning, and Statistics fields. They offer valid and successful solutions in various areas of fraud crime.
Techniques used for fraud detection are included in two main classes: statistical techniques and artificial intelligence. Examples of statistical data analysis techniques are:
- Data pre-processing techniques for detection, validation, error correction and filling missing or wrong data.
- Calculation of various statistical parameters such as averages, quantitative, performance metrics, probability distributions, and so on. For example, the average includes the average call duration, average number of calls per month, and average delay of bill payments.
- The model and probability distribution of various business activities either in terms of various parameters or probability distributions.
- Compute the user profile.
- Time series analysis of time-dependent data.
- Clusters and classifications to find patterns and associations between data sets.
- Match algorithms to detect anomalies in transaction or user behavior compared to previously known models and profiles. Techniques are also needed to eliminate false alarms, estimate risks, and predict future transactions or current users.
Some forensic accountants specialize in forensic analysis which is the procurement and analysis of electronic data to reconstruct, detect, or support fraudulent financial claims. The main steps in forensic analysis are (a) data collection, (b) data preparation, (c) data analysis, and (d) reporting. For example, forensic analysis can be used to review the activity of an employee's purchasing card to assess whether a purchase was transferred or diverted for personal use. Forensic analysis may be used to review invoice events for vendors to identify fictitious vendors, and this technique may also be used by franchisors to detect fraudulent or false sales reports by franchisees in franchising environments.
Management fraud is an intensive knowledge activity. The main AI techniques used for fraud management include:
- Data mining to classify, group, and segment data and automatically find associations and rules in the data that can indicate interesting patterns, including those related to fraud.
- Expert system to encode expertise to detect fraud in rule form.
- Pattern recognition to automatically detect suspected class (s), clusters, or suspicious behavior patterns, or to match the input given.
- Machine learning techniques to automatically identify fraudulent characteristics.
- A neural network that can study the suspicious patterns of the samples and use them later to detect them.
Other techniques such as link analysis, Bayesian networks, decision theory, and sequence matching are also used to detect fraud.
Video Data analysis techniques for fraud detection
Company
Younger companies in fraud prevention rooms tend to rely on systems that have been based around machine learning, rather than later incorporating machine learning into existing systems. These companies include FeaturespaceZensed, Feedzai, Stripe, Fraud.net, SecurionPay, Forter, Sift Science, Signifyd, Riskified, Experian and ThirdWatch. However, various security issues have been raised about how those solutions collect signals to detect fraud and how they are deployed.
Maps Data analysis techniques for fraud detection
Machine learning and data mining
Initial data analysis techniques are oriented to extract quantitative and statistical data characteristics. These techniques facilitate useful data interpretation and can help to gain better insight into the process behind the data. Although traditional data analysis techniques can indirectly lead us to knowledge, it is still created by human analysts.
To go beyond, the data analysis system must be equipped with a large amount of background knowledge, and can perform reasoning tasks involving that knowledge and data provided. In an effort to meet this goal, researchers have turned to the idea of ​​the field of machine learning. This is the source of a natural idea, because the task of machine learning can be described as changing the background of knowledge and the sample (input) into knowledge (output).
If the data mining finds a meaningful pattern, the data turns into information. New or valid information, and potentially useful information not only information, but also knowledge. One talks about finding knowledge, before it is hidden in large amounts of data, but it is now revealed.
Machine learning and artificial intelligence solutions can be classified into two categories: 'supervised' and 'unattended' learning. This method looks for accounts, customers, suppliers, etc. Which behaves 'amazingly' to produce a score of suspicion, rules or visual anomalies, depending on the method.
Whether a supervised or unattended method is used, note that output only gives us an indication of possible fraud. No stand-alone statistical analysis can ensure that a particular object is a fraudulent one. It can only show that this object is more likely to be cheat than other objects.
Supervised learning
In a supervised study, random sub-samples of all records are taken and classified manually as 'cheat' or 'non-fraud'. A relatively rare event such as fraud may need more than a sample to get a sizeable sample size. Records classified manually are then used to train supervised machine learning algorithms. After creating the model using this training data, the algorithm must be able to classify the new record as fraud or not cheat.
Controlled neural networks, blurred nerve webs, and a combination of net and neural rules have been explored and used to detect fraud in mobile phone networks and financial reporting fraud.
Bayesian learning nerve networks are implemented to detect credit card fraud, telecom fraud, automatic claim fraud detection, and health insurance scams.
Hybrid knowledge/statistics-based systems, where expert knowledge is integrated with statistical power, uses a series of data mining techniques for the purpose of detecting mobile clone fraud. In particular, the rules-learning program to uncover fraudulent behavior indicator of a large database of customer transactions is implemented.
Cahill et al. (2000) designed fraudulent signatures, based on data from fraudulent calls, to detect telecommunication fraud. To print a call for fraud, the probability is under an account signature compared to its probability under a fraud signature. Fraud signatures are updated in sequence, allowing fraudulent detection of events.
Link analysis understands different approaches. It deals with known con artists to other individuals, using a relationship of records and social networking methods.
Detection of this type is only able to detect fraud similar to that which has happened before and has been classified by humans. To detect a new type of fraud may require the use of unattended machine learning algorithms.
Unattended learning
Instead, unattended methods do not use labeled records.
Some important studies with unattended learning with regard to fraud detection should be mentioned. For example, Bolton and Hands use Peer Group Analysis and Breakeven Point Analysis applied to spending behavior in credit card accounts. Peer Group Analysis detects individual objects that start behaving in different ways than previously similar objects. Another tool Bolton and Hand develop for the detection of behavioral fraud is Break Point Analysis. Unlike Peer Group Analysis, Break Point Analysis operates at the account level. A break point is an observation in which an anomalous behavior for a particular account is detected. Both tools are applied to spending behavior in credit card accounts.
Also, Murad and Pinkas focus on behavioral changes for the purpose of fraud detection and present a three-level profiling. The three-level-profiling method operates at the account level and shows significant deviations from the normal behavior of accounts as potential fraud. To do this, the 'normal' profile is based on data without fraudulent records (semi-supervised). In the same field, Burge and Shawe-Taylor also use behavior profiles for the purpose of fraud detection. However, using repetitive neural networks for prototyping call behavior, unattended learning is applied.
Cox et al. combining human pattern recognition skills with automated data algorithms. In their work, information is presented visually by a domain-specific interface, combining human pattern recognition skills with automated data algorithms (Jans et al.).
See also
References
Source of the article : Wikipedia
0 Comments