The broader context of this study lies in the growing importance of securing power systems against increasingly sophisticated cyberattacks. As power grids and other critical infrastructures become more digitized, the potential attack surfaces expand, making robust anomaly detection systems crucial. Smart homes, which increasingly rely on IoT-based devices, are particularly vulnerable to cyberattacks. These attacks can have grave consequences, such as breaches of privacy, property damage, and even physical harm. The growing use of these devices by consumers highlights the importance of protecting them as a key sociotechnical issue. This work is based on the findings of a previous study, where attacks on ten smart home devices were simulated, identified visually in their power consumption data, and grouped based on similarities in their time series data. The primary objective of the present study is to develop and evaluate various classical machine learning models for the automated detection of cyberattacks using power consumption data. Given that many of the IoT devices in this study are consumer-focused, the practical relevance of these solutions for real-world smart home environments is emphasized. The initial time series data is used to create new features which are well suited for real-time monitoring. The performances of various Machine Learning models are examined, and the best models in terms of time and performance are presented. The results indicate that Extreme Gradient Boosting is particularly well-suited for real-time anomaly detection in power consumption monitoring systems, offering both high accuracy and efficiency across different device types.