The number of financial crimes has increased in modern times. Financial fraud must be identified to ensure that transactions remain secure and to maintain public trust. Machine Learning algorithms like Logistic Regression, Decision Trees, and Random Forests help us achieve this. The straight-forward method of Logistic Regression is used to identify linear relationships. Decision Trees, on the other hand, are more suitable to handle complex fraud patterns as they can capture non-linear relationships. Random Forests use numerous decision trees making them best suited for datasets which are at risk of overfitting.
Algorithms get evaluated by performance criteria such as F1 score,
accuracy and precision. Out of all the transactions labelled as fraud,
precision is the proportion of correctly identified fraud transactions. The
proportion of accurately classified transactions is known as accuracy. The
harmonic mean of precision and recall gives us the F1 score and a normalized
score of the model's performance is obtained by balancing.
Statistical hypothesis tests are applied for an accurate
comparison is used to compare. When three or more algorithms are to be compared
to check for a statistically significant difference between their means, we use
Analysis of Variance (ANOVA). This study aims to understand the effectiveness
of the ML algorithms in fraud detection by performing ANOVA tests on the
selected performance metrics. Here synthetic data set is generated and applied
statistical techniques for evaluation.
Author
(s) Details
W. Grace Shanthi
Kakatiya Institute of Technology and Science, Warangal, India.
Please see the book here:- https://doi.org/10.9734/bpi/mono/978-93-48859-10-5/CH5
No comments:
Post a Comment