Gas leaks at residential properties can lead to serious dangers like fires, explosions, and health issues from extended exposure. Conventional sensor-based systems often rely on thresholds but may fail under changing conditions or result in false alarms. This article introduces an innovative system for detecting gas leaks in smart homes using machine learning to boost accuracy, responsiveness, and reliability. The proposed setup combines real-time sensor data such as methane, LPG, carbon dioxide, temperature, and humidity with supervised algorithms like Random Forest, Support Vector Machine, and Gradient Boosting to classify normal and leak scenarios. The information is preprocessed with noise removal and feature standardisation techniques, followed by feature selection to optimise performance. The models are trained and evaluated on a labelled dataset from simulated smart environments and validated using metrics, including accuracy, precision, recall, and the F1-score. Experiments show that the Random Forest model detects over 98% of cases correctly with very few false alarms, doing much better than older systems that rely on set limits. The system can integrate into an Internet of Things architecture, providing immediate alerts and connecting to home automation for emergency responses, such as shutting gas valves or triggering alarms. This research paves the way for deploying AI-driven safety solutions in smart homes to proactively manage hazards and ensure occupant protection.