Extreme heat events in Canada pose a significant threat to public health, especially affecting vulnerable groups such as the elderly and individuals with pre-existing health conditions, who often spend considerable time indoors. Increasing indoor temperatures during heatwaves can worsen health problems and lead to heat-related illnesses. Accurate forecasting of indoor temperatures is thus crucial for informing public health strategies and mitigating the impacts of these extreme heat conditions.
This project addresses this challenge by predicting heatwaves by analyzing indoor temperature data collected from ecobee smart thermostats in North American households. We began our initiative by cleaning and merging extensive indoor and outdoor temperature datasets, ensuring the accuracy and consistency necessary for reliable analysis. To uncover the temperature dynamics and identify underlying patterns, we developed comprehensive time series features, including date, time, and seasonal indicators, facilitating a thorough examination of temporal trends in temperature data.
Utilizing the XGBoost machine learning algorithm, we are developing models to forecast indoor temperatures for the upcoming year, enhancing model accuracy with lag features incorporating historical temperature data. Our findings, visualized with Matplotlib and Seaborn, will contribute to developing near-real-time interactive, city-specific dashboards. Starting with Ottawa, these dashboards aim to understand the local heatwave patterns and their implications for public health officials and policymakers, providing essential insights. By leveraging sensor-based technologies and real-time data analysis, our research is an important component of early warning systems, offering vital information to improve response strategies and protect vulnerable populations during extreme heat events.
PI: Dr. Plinio Pelegrini Morita
Project members: Dr. Jasleen Kaur, Arlene Oetomo, Navneet Kaur, Aline Priscila de Souza