Votist

An API using FastAPI to screen transactions for fraud.

Endpoint(s)

Install

docker build -t votist . && docker run -p 80:80 --name=votist_container votist

Documentation

Random Forest Model

A simple random forest model was developed using data provided by Kartik Shenoy on Kaggle.

Feature Engineer

dob was converted to an age field, i.e. 1998-01-01 would return 23 (2021).
category was one-hot encoded.

Results

Confusion Matrix

[[257653    195]
 [   502    985]]

Classification Report

              precision    recall  f1-score   support

           0       1.00      1.00      1.00    257848
           1       0.83      0.66      0.74      1487

    accuracy                           1.00    259335
   macro avg       0.92      0.83      0.87    259335
weighted avg       1.00      1.00      1.00    259335

Accuracy Score

0.997312356604392

Performance

Votist was deployed using Docker (i7-8700K @ 3.70Ghz with 8 CPUs, 8GB of Memory) and load tested using Locust.

With 500 users the system handled 163 requests per second and responded in 21ms in the 50%ile range and 150 in the 95%ile range. Out of 376201 requests, 97 of them failed. Click here for the report.

With 1000 users the system handled 273 requests per second and responded in 250ms in the 50%ile range and 2300 in the 95%ile range. Out of 234095 requests, 2530 of them failed. Click here for the report.

Analysing the Votist container CPU usage using cAdvisor seemed healthy with an average of 90% usage.

We can improve performance by implementing a load balancing and scaling more instances of Votist to distrbiute the load. Additionally we can change our orchestrator from docker-compose to Kubernetes, to auto-scale our application for larger loads with netter monitoring and better CNIs.

ShaneNolan / votist