Temperature Prediction with Feature Engineering and Multiple Regression: Smart Buildings Hackathon Submission

This paper presents a methodology for the Smart Buildings Hackathon.The task was to predict air temperature sensor readings at various locations in a smart building.Using a dataset of device sensor and setpoint measurements, we developed and evaluated several regression models to forecast temperature values.Our bestperforming model, a Lasso regression with an alpha value of 0.003, achieved a root mean square error (RMSE) of 2.17, mean absolute error (MAE) of 1.75, and average Kullback-Leibler (KL) divergence of 0.74 on the validation set.We discuss the challenges encountered during the modeling process, particularly related to feature multicollinearity, and outline promising directions for future work to further enhance prediction accuracy.