<![CDATA[This study aims to implement TinyML based on Edge AI for anomaly detection on resource-constrained IoT devices in tropical environments in Indonesia. Tropical environments with high temperatures and extreme humidity often cause sensor data drift and other anomalies, disrupting the reliability of monitoring systems. This research was conducted through simulation without physical hardware to overcome cost and infrastructure limitations. The methods include collecting time-series datasets from open sources, augmenting tropical noise using Gaussian filter, training anomaly detection models (K-means and Autoencoder) on the Edge Impulse platform, INT8 quantization, and inference simulation using TensorFlow Lite Runtime in Python. Evaluation was performed on accuracy, inference latency, model size, and model robustness against tropical conditions. Simulation results show that the TinyML model achieved 89.4% anomaly detection accuracy, model size of 142 KB, and average inference latency of 28 ms. The model also demonstrated good resilience to tropical noise with only a 6.2% decrease in accuracy. The study concludes that TinyML based on Edge AI has high potential for implementation on low-power IoT devices in tropical environments. This research is expected to serve as a reference for developing intelligent monitoring systems in tropical regions of Indonesia, such as South Sulawesi.]]>
Copyrights © 2026
