Machine Learning in Finance

Traditional models follow rules and relationships specified by humans. Machine learning models instead find patterns in data on their own, often discovering complex relationships that would be hard to specify in advance.

Common applications include detecting fraudulent card transactions in real time, assessing creditworthiness using a wider range of data, predicting customer behaviour, automating document processing, supporting investment and trading decisions, and monitoring transactions for money laundering.

For example, a card issuer might train a model on millions of past transactions labelled as fraudulent or genuine, so it can flag suspicious new transactions within milliseconds.

Machine learning brings challenges. Some models are hard to explain, which matters when decisions such as refusing credit must be justified to customers and regulators. Models can learn biases from historical data, leading to unfair outcomes. They can overfit past data and fail when conditions change. Regulators increasingly expect firms to govern, validate and explain the models they use.