Last week, Accenture Consulting published a presentation on how banks can utilize machine learning to draw quicker insights from their data. Machine learning is the specific application of artificial intelligence (AI) in which computers can learn without being explicitly programmed to do so.
Machine learning begins with an identified data set that will be used to “train” the computer. If the goal is for the computer to make a judgment based on data, then you would also need historical data that can be matched with correct answers. Using historical data as a training guide, the computer can now be programmed to go through real world data sets searching for similar patterns and making predictions. Here’s the interesting part: as the algorithm goes through more and more data sets, it improves and adjusts itself based on whether or not its predictions became true; it learns on its own.
There are two main kinds of machine learning: supervised and unsupervised. Supervised machine learning is when you know the output variable and are trying to predict future outputs. The example Accenture gave was credit default risk. Presumably, you could program an algorithm to go through historical data on individuals who have defaulted on debt and identify key metrics that could be a marker for default risk. The algorithm could then predict future cases of individuals defaulting and adjust itself based on whether or not those individuals actually defaulted.