Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{207063,
        author = {Karri Thrishank and Ch K Rupesh Kumar and S Virendra and S Dinesh Kanna and N Sushma},
        title = {An IoT-Enabled Microclimate Monitoring System with Online Machine Learning for Real-Time Temperature Forecasting},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {2},
        pages = {4256-4264},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=207063},
        abstract = {Reliable microclimate monitoring is difficult in regions without dense meteorological infrastructure, and commercial weather services often describe conditions in a distant city rather than at the specific site of interest. This gap motivated Tempo, an end-to-end IoT and machine learning system built to forecast short-term temperature at the Anil Neerukonda Institute of Technology and Sciences (ANITS), Andhra Pradesh, India. An ESP32 microcontroller fitted with a DHT22 sensor records temperature, humidity, and heat index at regular intervals; a FastAPI backend hosted on Render ingests these readings and, whenever the sensor is unavailable, substitutes atmospheric data from WeatherAPI.com so the pipeline keeps running. Forecasts are produced by a Stochastic Gradient Descent Regressor (SGDRegressor) that updates incrementally through scikit-learn's partial_fit() method, allowing the model to track diurnal and seasonal patterns without full retraining. A scheduler recomputes rolling Root Mean Square Error (RMSE) every six hours and initiates corrective retraining whenever the error exceeds 1.5°C. Over a seven-day evaluation window, the deployed model achieved an RMSE of 0.734°C, an MAE of 0.521°C, and an R² of 0.961 — performance close to that of static batch models while retaining continuous adaptability. A parallel offline comparison using PyCaret found that Gradient Boosting Regression reaches a marginally lower RMSE of 0.541°C but cannot support incremental updates, making it less suited to a continuously evolving sensor stream. Predictions and performance logs are stored in a Supabase PostgreSQL database and presented through a React 18 and Vite dashboard that combines Gemini-generated explanations with configurable email alerts.},
        keywords = {IoT, online machine learning, SGDRegressor, temperature forecasting, ESP32, FastAPI, Supabase, drift detection, partial_fit, microclimate monitoring},
        month = {July},
        }