Fraud Analytics

What Are Fraud Analytics?

Fraud analytics are technical methods to detect and monitor fraud, which occurs when people intentionally act secretly to deprive another of something of value. Fraud analytics can take place either before the transactions are completed (fraud prevention) or after they occur (fraud detection). Fraud analytics help organizations reduce costs associated with fraud.

Financial fraud can be corporate, such as when a financial statement is falsified, when hedge funds falsely report returns, or when stock markets manipulation jeopardizes regulation compliance. Fraud can also happen in healthcare and insurance and through methods such as identity theft (credit cards), money laundering, and tax evasion.

Hedge fund returns manipulation is more prone to fraud, due to fewer regulations. It results from misbehavior when managers have discretion in valuing illiquid investments or commit outright fraud.

The most commonly used techniques in fraud analytics are artificial intelligence (AI), machine learning, deep learning, and statistical analysis. You can apply these techniques with MATLAB® to detect which banking transactions are potentially fraudulent.

  • Machine learning (supervised or unsupervised) gives you an indication of the likelihood of fraud with a high degree of accuracy at detecting anomalies.
    • In supervised learning (regression, classification), historical transactions are labeled as fraudulent or genuine. Then these records are used to train an algorithm that infers a function able to classify a future transaction as either legitimate or not. A typical example in regression is to predict the amount of fraud.
    • Unsupervised learning does not require historical observations to be labeled as fraudulent or non-fraudulent and is useful for companies that don’t have historical fraud data available.
  • Data mining and pattern recognition are used to detect meaningful patterns or trends among the data that relate to fraud.
  • Statistical analysis consists of an analytical framework to calculate statistical parameters to identify outliers that could indicate fraud patterns.

Finally, Benford's law could be used as an indicator to detect fraud. Other indicators used for fraud analytics include return-based and text-based.

For more on fraud analytics, see Statistics and Machine Learning Toolbox™, Deep Learning Toolbox™, and Text Analytics Toolbox™.

See also: AI in finance, predictive analytics, Machine Learning in Finance (9 videos), risk management, statistical arbitrage