This paper presents a complete design, implementation, and evaluation of a forest fire detection system built on Internet of Things (IoT) principles and presented in IEEE-style format. The proposed system integrates a network of low-cost sensor nodes (temperature, humidity, smoke, CO), microcontroller-based edge processing, and a multi-tier communication architecture (local gateway → cloud). Detection is performed using a rule-based thresholding stage followed by a lightweight machine learning classifier on the gateway for improved false alarm suppression. The system supports real-time alerts (SMS/Push), geolocation tagging, and dashboard visualization. We evaluate the system through controlled experiments and simulations and report metrics including detection accuracy, false alarm rate, detection latency, energy consumption, and network overhead. Results show that the combined threshold+ML approach reaches high detection accuracy while maintaining low energy use, making it suitable for wide-area forest deployments.