you are onto a simple task Muhammad Ghalib Mirza , tree based learning algorithms does really on tabular data,you can try decision tree to solve your problem,if it doesn't work well on your data then i believe random forest will do pretty well,if your datasets dimension is high then there is high probability that your ml model will overfit,you can try simple to advance feature selection techniques using multivariate statistics techniques like clustering to select important features or PCA to reduce the dimensionality of your dataset,then after reducing dimension you can feed new feature to random forest and i am optimistic that it will do pretty well job on learning how to label your data.if most of your features are categorical then you can also try "catboost algorithm" which is a categorical boosting learning algorithm and has the ability to handle categorical features well,thanks

Need methods of intelligentdata analysis to detect and prevent it

The proses of Audit in tha information is important, methods in the SGBD

A simple but quite efficient method may be to perform a semi-supervised learning, for labeling your data and also developing a fraud detection model in this context.

The process is to:

1) manually label a subset of your data as fraud/not or 1 & 0 respectively,

2) train your desired model with the label data - supervised learning.

3) Divide the remaining data set into two (one for re-training and the other for test)

4) test the trained model on the data for re-training and get the corresponding labels.

5) add the newly labeled data to the manually labeled data.

6) Retrain your model on the new data obtain in step (5).

7) Re-test your model on the test data.

This should produced a viable solution.

Note: the entire model development phase is executed at your discretion and expertise

Hi, I think you are working with text mining, if yes, so you are need to built the lexicon which content on the list of word with label fraud and non fraud with the weights for each single word.

Thank you everyone!

Victor Adeyemo thats what i have in mind except the re-training par, but isnt the step 4 comes before step 2?

And i think these steps are very much similar to the concept of Online Learning. Isn't it?

Hi, In my understanding, Decision tree will work best for this problem case.

Sorry this is coming late!

Muhammad Ghalib Mirza

Step 2 develops a trained model (call it model 1) using the manually labelled data.

In Step 4, model 1 will be used to predict the classes for the unlabelled data which was initially set aside for re-training.

Thus, the sequence I highlighted is correct

@victor i understand your point.

But i think we are slipping from the actual question which is "How to label the data" point1.

Is there something like automated feature engineering? How credible is it?

Muhammad Ghalib Mirza

The actual question is "How to label the data?", I gave a response solution via labeling data using a handful of labeled rows of data by a domain expert to label the largely unlabeled part of the data.

Automated feature engineering is possible . I am yet to work on it and I can not speak about its credibility.

Regards.