Real-time Fraud detection web application
In the first part the machine learning part is going to be described. This fraud detection project notebook could be found here.
This fraud detection process has been implemented with two algorithm, with Random Forest and CNN and accuracy of 99% on test data.
Because of more simplicity of random forest model; in this project this model is employed to check if sent data is fraud or not!
Cassandra
The database for this real-time web application is Cassandra. This is the class for creating connection and key-space:
The data model for result data is like this:
WEB
This part of project has been built with Fastapi, this project has two endpoint to be called for checking data for fraud.
RESTfull api
This method is just simple posting data and getting back the results:
The input data:
-0.54839782, -0.35184723, 0.17023924, -3.10939288, 0.909047 ,
3.53073928, -1.49890445, 1.37163425, -2.58719536, 0.75480169,
-0.59832237, -1.24748237, 0.56082062, -0.1370649 , 0.78843122,
0.1627786 , 0.08075539, 0.97238009, 0.14502653, -0.2140704 ,
0.01214456, 0.11382656, -0.18937492, 0.99338122, 0.13761257,
-0.10732711, 0.00999518, 0.03007495, 7.5
As it can be seen post method route has done its job right!
Web-socket
This feature is very awesome, it will classify the data in real-time and turn back the results with calling button. But to support web socket in Nginx we need to make serveral changes:
The config file is here.
The input data:
-0.54839782, -0.35184723, 0.17023924, -3.10939288, 0.909047 ,
3.53073928, -1.49890445, 1.37163425, -2.58719536, 0.75480169,
-0.59832237, -1.24748237, 0.56082062, -0.1370649 , 0.78843122,
0.1627786 , 0.08075539, 0.97238009, 0.14502653, -0.2140704 ,
0.01214456, 0.11382656, -0.18937492, 0.99338122, 0.13761257,
-0.10732711, 0.00999518, 0.03007495, 7.5
As it can be seen the length of data must be 29.
Checking Cassandra
By calling the root API the number of data is going to e shown that has been saved into Cassandra:
Yeah there is 2 item that we tested!
