Research Symposium

BIO

Layhan Mishra is a first-year computer science student from Olathe, Kansas. He is interested in research topics related to computer science, including, but not limited to, machine learning, cybersecurity, and artificial intelligence. Layhan hopes to eventually work as a software engineer or pursue higher education following graduation.

Using Machine Learning to Recognize Attacks on Power Grids

Abstract

Many hackers perform cyberattacks on power grids to reduce their utility bills. This research was conducted to determine which machine-learning models are most effective in detecting such cyberattacks on power grids. Given a dataset from an Irish power company with information on several users’ power usage and whether they artificially reduced their utility bills, multiple machine-learning models were trained on a large portion of the dataset and then tested on a smaller portion. The models were then evaluated on 4 metrics: accuracy, precision, recall, and F1 score. Because of the variety of statistics evaluated and the variety of machine learning models, there is no clear-cut best-performing machine learning model. However, taking all data into account, there were three models that performed the best: the random forest, decision tree and CNN. Out of these three, the random forest performed the best consistently across all metrics. However, it should be said that the decision tree and CNN also detected attacks at a very high rate and could be better than the random forest for different instances of this scenario (different power companies, cities, and power grids). For this particular scenario, any of these three could realistically be used to detect cyberattacks on power grids with the random forest classifier being the best.

Keywords: Machine learning, power grids