Table of Contents
Question
You work for a credit card company that’s trying to do a better job identifying fraudulent transactions. So your team uses unsupervised learning to create clusters of transactions that are likely to be fraudulent. The machine identified that when customers are buying electronics it’s much more likely to be a fraudulent transaction. So you use this model for your new fraud detection system. Then customers started to complain that they couldn’t use their credit cards to purchase any electronics. What is the challenge with your model?
A. You used too much data to train the algorithm how to make predictions.
B. You used unsupervised learning when you should have used supervised learning.
C. You overfit the model to the data, the added complexity made it difficult to manage the system.
D. You underfit the model to the data, the simple rule made too many inaccurate predictions.
Answer
D. You underfit the model to the data, the simple rule made too many inaccurate predictions.
Explanation
This is an interesting question. It seems that you are using unsupervised learning to create clusters of transactions that are likely to be fraudulent, based on some similarity measure. Unsupervised learning is a type of machine learning that does not use any labels or targets to guide the learning process, but rather tries to find hidden patterns or structures in the data.
One of the challenges with your model is that it may be too simplistic and generalized to capture the complexity and diversity of the real-world data. By using a single rule that when customers are buying electronics it’s much more likely to be a fraudulent transaction, you are ignoring other factors that may influence the fraud probability, such as the amount, location, time, frequency, or history of the transactions. This may lead to many false positives, i.e., transactions that are flagged as fraudulent but are actually legitimate. False positives can cause customer dissatisfaction, loss of revenue, and damage to reputation.
Another challenge with your model is that it may not be able to adapt to new or changing patterns of fraud. Fraudsters are constantly evolving their strategies and techniques to evade detection, so a static model based on historical data may not be able to catch them. Unsupervised learning models need to be updated frequently with new data and feedback to maintain their accuracy and relevance.
Therefore, a possible solution to improve your model is to use supervised learning instead of unsupervised learning. Supervised learning is a type of machine learning that uses labeled or target data to train the model how to make predictions. For example, you can use a dataset of transactions that are labeled as fraudulent or non-fraudulent, and train a classifier that can predict the fraud probability of new transactions based on various features. Supervised learning models can be more accurate, specific, and robust than unsupervised learning models, as they can learn from the feedback and correct their mistakes.
To answer your question, the best option among the choices is D. You underfit the model to the data, the simple rule made too many inaccurate predictions. Underfitting means that the model is too simple or has too few parameters to capture the complexity or variability of the data. Underfitting can result in high bias and low variance, meaning that the model makes many errors and does not change much with different data sets.
Reference
- Fraud Detection applying Unsupervised Learning techniques | by Mauro Krikorian | SOUTHWORKS | Medium
- [2208.01203] Unsupervised quantum machine learning for fraud detection (arxiv.org)
- Fraud Detection Algorithms | Fraud Detection using Machine Learning (intellipaat.com)
- Unsupervised Machine Learning to Reduce False Positives in Fraud Detection – DataVisor
The latest Generative AI Skills Initiative certificate program actual real practice exam question and answer (Q&A) dumps are available free, helpful to pass the Generative AI Skills Initiative certificate exam and earn Generative AI Skills Initiative certification.