Machine learning (ML) is impacting financial services in many different ways. For example, it is automating routine lending decisions, it is proving useful in fraud detection and anti-money laundering, and it is making risk management and derivatives trading more efficient. This research report:
• Explains the tools commonly used in ML
• Identifies some of the challenges ML poses for financial institutions
• Formulates key questions that senior management should ask in regards with ML applications
In our discussion we cover supervised, unsupervised, and reinforcement learning. In addition, we also discuss one of the most common applications of ML – Natural Language Processing.
In a supervised learning model, the data scientist specifies an objective and supervises the learning to ensure that the objective is met. The objective can be the forecast of something (e.g., the value of a house) or the classification of data (e.g., estimating which transactions are likely to be fraudulent and which are not). In unsupervised learning, no objective is specified by the data scientist. The algorithm looks for patterns in the data. Reinforcement learning is concerned with discovering actions (e.g., trading strategies) that maximize a certain objective function (e.g., risk-adjusted returns) by trial-and-error interactions with the environment. Natural language processing is an important application of ML concerned with processing and analyzing human language data by computers.