AIRA: An IOT-Enabled Machine Learning System for Real-Time Stress Detection

We trained the system with a Random Forest classifier on the WESAD benchmark dataset. When we put it to the test, AIRA hit 95.2% accuracy at distinguishing stress from no-stress, and 88.3% for sorting into all three levels. We also ran it with real people wearing the devices for two weeks and got a mean accuracy of 90.4%, checked against ground-truth labels. The system is quick, too—on average, it reports stress in under 300 milliseconds, and even when 500 devices connect at once, it Stress is everywhere these days, and it’s not just an annoyance—it’s a real health problem. We know it’s linked to things like heart disease, trouble with the immune system, memory problems, and even mental illnesses. But here’s the thing: most people have no real way to monitor their stress as it happens, especially outside a hospital or doctor’s office. That’s where our project, AIRA (Adaptive Intelligent Real-time Analysis), comes in. We built an end-to-end system that uses IoT and machine learning to keep tabs on stress in real time. It works by pulling data—heart rate, body temperature, and motion—from off-the-shelf smartwatches. This data gets sent up to the cloud, where our machine learning pipeline crunches the numbers and spits out stress levels, daily health insights, and even sends alerts to caregivers if needed. AIRA sorts stress into three categories: Lowstill responds in less than two seconds. In this paper, we walk through how AIRA is built, how the data gets processed, the machine learning approach, our results, and where we think this technology goes next. AIRA’s big promise is making stress monitoring practical and available for anyone—not just people in clinics or hospitals. It’s a real step forward in affective computing and preventive digital health.