Insurance fraud claims analysis project
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore initial commit Mar 15, 2017
Color.py Added more files May 13, 2017
Fields.ods
LICENSE Create LICENSE Jun 5, 2018
README.md Added create database in excel format Oct 10, 2018
cleaning.py
createDatabase.py
eda.py updated project for final release Jun 5, 2018
insurance.pdf
machine_learning.py updated project for final release Jun 5, 2018
ppdm.py updated project for final release Jun 5, 2018

README.md

Fraud Analysis

A project on fraud analysis of insurance claims.

Requirements

- Python
    - mimesis (elizabeth is no longer available)
    - faker
    - matplotlib
    - sqlite3
    - numpy
    - pandas
    - sklearn
    - pydotplus
    - openpyxl

Phases

Create Database

Create the database with meaningful random data.

python createDatabase.py

Cleaning the data

Cleaning invalid data/rows that does not meet a defined criteria. Fills in data using samples or a mean/median.

python cleaning.py

EDA

Exploratory Data Analysis - understand the data and data types as well as some statistics and graphing to see the distribution, correlation, anomalies and outliers of the data.

python eda.py

PPDM

Privacy Preserving Data Mining - suppress, generalize, anatomization, perturbation, categorize, k-anonymity is done in order to preserve privacy so that sensitive attributes cannot identify a person without having the entire dataset. This makes the data safer in an instance the data is leaked, it makes it harder to impersonate someone.

python ppdm.py

Machine learning

Machine leaning was used to detect fraudulent insurance claims. This uses a simple decision tree classifier and was trained with 70/30 train/test ratio. The accuracy of the prediction was ~99% with 73117 training elements and 18280 testing elements. The tree can be seen in insurance.pdf.

python machine_learning.py