AI and Fraud Prevention: Part I

Sharing knowledge is a two-way street. I advise and equally I like to learn. With all the fuss about AI I wanted to know more about its potential for counter-fraud, so I tracked down a recommended AI Subject Matter Expert specialist to ask him some questions. He asked me about fraud and working together we produced these articles.

It is a four part series, providing bite size chunks of information that are easier to consume.

Article One: a glossary of AI terms. If you are not an AI expert, this is more than enough to familiarize yourself with right now. Article Two will be released next Friday and answers the question: What is it about AI that can cut through the fog of complex fraud prevention? But first things first…

Glossary of Terms:

Artificial Intelligence (AI): The capability of a machine to imitate intelligent human behaviour.

Machine Learning (ML): An application of artificial intelligence that enables systems to learn and improve from experience without explicit programming.

Supervised Learning: A type of ML where the model is trained on labeled data, i.e., data with known outcomes.

Self-supervised Learning: An ML technique where the model learns from unlabeled data and identifies patterns on its own.

Reinforcement Learning: An ML method where a model learns to make decisions by experiencing consequences in a dynamic environment.

Neural Network: A computing model inspired by the human brain’s biological neural networks. It is designed to recognize patterns.

Deep Learning: A type of ML that uses neural networks with many layers (deep neural networks) to analyze several factors with a structure like the human brain’s neural network.

Feature Selection/Extraction: The process of selecting the most relevant input variables for training a machine learning model.

Overfitting: A modeling error in ML that occurs when a function is too closely fit to a limited set of data points, causing poor predictive performance on new data.

Underfitting: The opposite of overfitting; when a model is too simple to accurately represent the complexity of the data, leading to low accuracy.

Bias-Variance Tradeoff: A problem in ML where increasing the bias will decrease the variance, and vice versa. It represents the balance needed between fitting the data well and generalizing to new data.

Training Set: The portion of data used to fit the model in supervised learning.

Test Set: The portion of data used to assess the performance of a fitted model, providing an unbiased evaluation.

Cross-Validation: A technique used to evaluate the effectiveness of ML models by partitioning the original dataset into a training set to train the model, and a test set to evaluate it.

Algorithm: A set of mathematical procedures that ML uses for learning from data.

Hyperparameters: The external configurations of the model that the learning algorithm does not learn but are set beforehand.

Classification: A type of supervised learning where the output is a category.

Regression: A type of supervised learning where the output is a real or continuous value.

Anomaly Detection: The identification of rare items, events or observations which raise suspicions by differing significantly from most of the data.

Precision and Recall: Precision is the number of correctly identified positive results divided by the number of all positive results. Recall is the number of correctly identified positive results divided by the number of all samples that should have been identified as positive. They are often used in tandem to evaluate the effectiveness of ML models.