False Data Injection Attacks (FDIAs) pose significant threats to the reliability and stability of smart grid operations by manipulating measurement data to mislead monitoring and control systems. This research presents a lightweight framework for real-time detection of FDIAs that balances computational efficiency with predictive accuracy. The proposed pipeline integrates four core components: data preprocessing, dimensionality reduction using Locally Linear Embedding (LLE), classification through an Extreme Learning Machine (ELM), and deployment via a Gradio-based interactive interface. Data preprocessing ensures noise reduction and consistency in raw sensor inputs, while LLE captures essential nonlinear relationships and reduces feature complexity. The ELM classifier, chosen for its fast training speed and strong generalization capability, effectively discriminates between genuine and manipulated data. To enhance usability, the framework is deployed on a Gradio interface, providing an accessible platform for real-time monitoring and decision support. Experimental results on a synthetic Smart Grid FDIA dataset demonstrate the framework’s ability to achieve high detection accuracy with minimal computational overhead, validating its suitability for practical, large-scale grid environments. By combining accuracy, speed, and user-friendly deployment, this study contributes a scalable solution to strengthen cyber-physical security in smart grids against evolving FDIA threats.