I would like to use 6 years data for 100 compoanies i.e., 600 sample data. Is this data okay for the ANN study?
I need a clear expression.
First you should define a clear goal for mining your data. What is your goal of analyzing this data and another point you should consider is your data details. Then you can choose the proper data mining technique to mine your data. Good luck
This sounds like a trick question. An analogy from politics mights help: how can you detect when a politician is lying? Answer: his or her mouth is moving. But seriously, it depends what you mean by fraud. Many financial practices that are in current use would have been considered fraudulent 50 years ago. Examples include tax havens and financial speculation. The fractional reserve banking system is a form of fraud.
Disclaimer: All comments are my own and do not reflect the views of my employer.
What attributes does your data have? And what proportion of the 600 points are fraud or legal practices?
It is definitely possible to detect fraud using data mining and machine learning techniques (not sure I've seen it done using ANN specifically but I don't see why it would be impossible), but only if your data has both enough attributes and enough details for an algorithm to learn to spot the differences, and if it has both positive and negative examples.
Problem with ANN, in my opinion, is that unless you have some previous literature on fraud detection networks, then you're starting by taking a shot in the dark on the structure and size of the network and type of transfer function. That said, there are ANN software libraries out there, so you could use one to implement and train several networks and weed out some unpromising options early.
I share with Graham the need to clearly define the financial fraud in context. It will also be good to know whether you intend to look at it from public or private organization point of view plus information on specific industry for the financial fraud is key for idea generation.
Good Luck.
Agree with Graham and Frank. Another issue is that data analytics may provide a limited picture about some frauds that fit your algorithms but you may never know about the frauds that don't.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling