In this paper, we detail a machine learning-DNN centric project aimed at constructing a robust system capable of identifying power theft and potential electrical anomalies. Leveraging a dataset comprising labeled non-fraudulent power consumption records with 11 features, the target variable is derived. Through supervised machine learning techniques, encompassing Logistic Regression, Random Forest and DNN algorithms, the project undertakes data pre-processing, feature engineering, model training, evaluation, and optimization. Performance evaluation highlights the Random Forest model as the top performer, achieving a Root-Mean-Square error of ~ 0 % and an accuracy greater than 95 % consistently. Additionally, the report delves into the future recommendation techniques to improve the fraud detection model.
Securing Mobile Connectivity: A Data-Driven Approach for Detection of Electricity Theft and Anomalies in 4G and 5G Telecommunications Base Stations.
1 Views
2 Downloads