Detection of fraude Electricity and Gas consumption for Société tunisienne de l'électricité et du gaz (STEG)

Fraud detection in electricity and gas consumption powered by machine learning

Results summary

• Data insights

Model summary

The best model: performance

The best model is an XGBoost which consideres all engineered features

ROC curve	Confusion matrix

Recall: 0.34 Precision: 0.59 ROC AUC: 0.9 F-beta: 0.58

                Classification Report:
--------------------------------------------------
              precision    recall  f1-score   support

         0.0       0.96      0.99      0.97     95945
         1.0       0.59      0.34      0.43      5674

    accuracy                           0.95    101619
   macro avg       0.77      0.66      0.70    101619
weighted avg       0.94      0.95      0.94    101619

Project Workflow

Selected metric

• F1 is the metric of choice as a balance between precision and recall with the weight on recall. The model should detect as many frauds as possible by keeping the number of false alerts low.

Hypotheses generated based on data exploration

1. Households in regions where fraud occurs more frequently than average are more likely to commit fraud themselves (check from the slide of Katja and correct)

2. Households with large variations in energy or gas consumption are more likely to commit fraud

Data source

• Data are retrieved from a Zindi competition (source: https://zindi.africa/competitions/fraud-detection-in-electricity-and-gas-consumption-challenge/data)

• One file named "train.zip", containing two files: i) one contains client data (named "train_client.csv"); ii) one contains billing history from 2005 to 2019 (named "train_invoice.csv").

Data description

Click to unfold

Client dataset

Feature name	Description
Client_id	Unique identifier for client
District	District number associated to the location of a client
Client_catg	Class the client belongs to; different class groups: 11, 12, 51
Region	Region number associated to the location of a client
Creation_date	Date when a client became customer of STEG
Target	Category indicating fraudolent client when is equal to 1 and no fraudolent client when is equal to 0

Invoice dataset

Feature name	Description
Client_id	Unique identifier for client
Invoice_date	Date of issue of the invoice relating to a given quarter due date for a client
Tarif_type	Type of way a client is charged for its electricity and gas copnsumption by the company STEG; each type is associated to a unique number
Counter_number	Serial number identifing the counter materialized by a series of numbers written on the counting device and unique for each client
Counter_statue	Working status of the device measuring the amount of energy consumed by a client, ranging from 1 to 5 (check together the values in that column?)
Counter_code	Registration number identifing the device made up of three digits located within the serial number of the meter
Counter_coefficient	Coefficient used to convert the raw meter readings from the meter into actual consumption values
Consommation_level_1	Consumption level 1: it is less than 2.400 kwh per year corresponding to the cost of 181 millimes per kwh (source: https://kapitalis.com/tunisie/2022/05/12/tunisie-les-nouveaux-tarifs-de-la-steg/)
Consommation_level_2	Consumption level 2: it is between 2.401 and 3.600 kwh per year corresponding to the cost of 223 millimes per kwh (source: https://kapitalis.com/tunisie/2022/05/12/tunisie-les-nouveaux-tarifs-de-la-steg/)
Consommation_level_3	Consumpution level 3: it is between 3.601 et 6.000 kwh per year corrersponding to the of 338 millimes per kwh (source: https://kapitalis.com/tunisie/2022/05/12/tunisie-les-nouveaux-tarifs-de-la-steg/)
Consommation_level_4	Consumption level 4: it exceeds 6.000 kWh per year corresponding to the cost of 419 millimes per kwh. https://kapitalis.com/tunisie/2022/05/12/tunisie-les-nouveaux-tarifs-de-la-steg/
Old_index	Old counter meter reading
New_index	New counter meter reading
Months_number	Number of the month (where the meter reading was taken?);
Counter_type	Type of device measuring the amount of energy consumed; ELEC = measuring electricity consumption; GAZ = measuring gas consumption

---

Set up

macOS type the following commands :

• For installing the virtual environment you can either use the Makefile and run make setup or install it manually with the following commands:

  make setup

  After that active your environment by following commands:

  source .venv/bin/activate

Or ....

• Install the virtual environment and the required packages by following commands:

  pyenv local 3.11.3
  python -m venv .venv
  source .venv/bin/activate
  pip install --upgrade pip
  pip install -r requirements.txt

WindowsOS type the following commands :

• Install the virtual environment and the required packages by following commands.

  For PowerShell CLI :

  pyenv local 3.11.3
  python -m venv .venv
  .venv\Scripts\Activate.ps1
  pip install --upgrade pip
  pip install -r requirements.txt

  For Git-bash CLI :

  pyenv local 3.11.3
  python -m venv .venv
  source .venv/Scripts/activate
  pip install --upgrade pip
  pip install -r requirements.txt

  Note: If you encounter an error when trying to run pip install --upgrade pip, try using the following command:

  python.exe -m pip install --upgrade pip

Usage

In order to train the model and store test data in the data folder and the model in models run:

Note: Make sure your environment is activated.

python example_files/train.py  

In order to test that predict works on a test set you created run:

python example_files/predict.py models/linear_regression_model.sav data/X_test.csv data/y_test.csv