ICACT20220012 Question.1
Questioner: spcho@hanmail.net 2022-02-15 오후 11:47:33 |
ICACT20220012 Answer.1
Answer by Auhor patrick.hosein@sta.uwi.edu 2022-02-15 오후 11:47:33
Chrome  Click!! |
| Can you explain in the procedure how you defined TP, TN, FP, and FN in the CTRT model? |
A True Positive (TP) is a call flagged as fraud that is actually fraud, while a True Negative (TN) is a call flagged as legitimate that is actually legitimate. A False Positive (FP) is a legitimate call flagged as fraud, which can result in poor customer service if the is incorrectly terminated. A False Negative (FN) error is a fraudulent call flagged as legitimate, which can lead to revenue leakage for the company. In our model, we attached a cost of 1 to a FN error and 0.1 to a FP error. |
ICACT20220012 Question.2
Questioner: spcho@hanmail.net 2022-02-15 오후 11:55:45 |
ICACT20220012 Answer.2
Answer by Auhor patrick.hosein@sta.uwi.edu 2022-02-15 오후 11:55:45
Chrome Click!! |
| It is stated that the CRT decision tree model achieved acceptable predictive accuracy. It is not clear whether the CRT model is a reliable model. Can you be more specific? |
The model's accuracy was defined aa the number of correct classifications predicted divided by the total number of predictions, giving over 80% for both the traditional accuracy-based CART model and the cost-sensitive CART model. An SVM approach was also explored, where roughly 79% predictive accuracy was achieved - thus the CART approach was selected as more reliable. |
ICACT20220012 Question.3
Questioner: tomayoon@ieee.org 2022-02-16 오전 2:24:32 |
ICACT20220012 Answer.3
Answer by Auhor patrick.hosein@sta.uwi.edu 2022-02-16 오전 2:24:32
Chrome Click!! |
| First of all, thank you for the good research paper and presentation contents. I would like to know more about the research environment used in this study, and also in detail about the equipment & operating systems, development languages and software libraries you used. |
Work was conducted using the Windows 10 version 20H2 operating system. We used R Statistical Software, relying on several ML packages such as FSelector, caret, mlbench. |
ICACT20220012 Question.4
Questioner: vasaka.vis@mahidol.edu 2022-02-16 오전 3:12:59 |
ICACT20220012 Answer.4
Answer by Auhor patrick.hosein@sta.uwi.edu 2022-02-16 오전 3:12:59
Chrome Click!! |
| The topic is interesting.
On slide No.9, you mentioned about 18 call variables/features, what are they?
Features in the blue box on the right hand-side contains only 7+2 features.
So, finally in your CART model, you use these 7 features? 9 features?
Moreover, when comparing with the existing work for example [Hiuchen Li et. al., 2018], could you improve the detection accuracy?
|
Those listed in the blue box are examples of main features selected. 18 features were considered for the CART model: Account Number, Server Number, Dialling Number, Dialled Number, Call Date, Call Duration, Call Time Plus Duration, Total Charges, Lead Dialled Number, Grouped Consecutive Dialled Number, Number of Calls Dialled Daily, Number of Calls per Group Daily, Daily Average Number of Calls To Destination, Daily Average Revenue to Destination,Daily Average Minutes of Use To Destination, Week Day, Time of Day, and TOS Category, as listed in Section IV of the paper.
We expect that accuracy can be improved by resampling the training/testing data or by including features that were not considered in this development |
ICACT20220012 Question.5
Questioner: namacabale@gmail.com 2022-02-16 오전 3:34:40 |
ICACT20220012 Answer.5
Answer by Auhor patrick.hosein@sta.uwi.edu 2022-02-16 오전 3:34:40
Chrome Click!! |
| Dear Author, you have a good research. Is there an existing technology almost similar to your proposal that has been deployed in real world environment? What does it lack that your work specifically addressed? Thank you and stay safe.
|
Related works have explored ML solutions for telecommunications fraud detection and prevention. Our research aims to highlight the benefit of employing cost-sensitive analyses in these models. It was found that a cost-sensitive model may sacrifice some predictive accuracy but has the potential to result in tangible benefits to the company, especially in terms of cost |
ICACT20220012 Question.6
Questioner: honhonzone@gmail.com 2022-02-18 오전 12:38:42 |
ICACT20220012 Answer.6
Answer by Auhor patrick.hosein@sta.uwi.edu 2022-02-18 오전 12:38:42
Chrome Click!! |
| This is a good topic. Can this method be used to detect financial fraud? |
Absolutely, this approach can be adopted for a variety of classification problems. |
