This project is focused on using machine learning techniques to identify fraudulent electricity consumption in a population in China. The dataset used for this project contains information on the electricity consumption of individuals in the population and whether or not they have been labeled as fraudulent.
The dataset used for this project contains the following features:
id: unique identifier for each individualdate: each date colum represents the day the data was preleved in.label: whether or not the individual has been labeled as fraudulent (1 = fraud, 0 = not fraud)
The dataset contains 10,000 observations.
Before building any models, we performed exploratory data analysis to gain insights into the dataset. We visualized the distribution of the consumption feature and found that it was positively skewed, indicating that the majority of individuals consume less electricity than the average.
We also looked at the distribution of the label feature and found that there were significantly more non-fraudulent individuals than fraudulent individuals in the dataset.
We used several machine learning models to predict whether or not an individual was fraudulent based on their electricity consumption. The models used were:
- K-nearest neighbour
- Logostic Regression
- Support Vector Machines (SVM)
- XGBOOST
In this project, we used machine learning techniques to identify fraudulent electricity consumption in a population in China. We found that the Random Forest model performed the best in predicting fraud, with an accuracy of 90%. This could be used to help identify individuals who are committing fraud and take appropriate action.