Department
Computer Science and Cybersecurity
Document Type
Poster
Abstract
Modern building management systems rely on continuous streams of sensor data for tasks such as occupancy prediction, energy optimization and system control(Zhang et al.; Ahmad et al.). Real world scenarios have seen sensor failures significantly degrade the performance of most Machine Learning models deployed in buildings(Khan et al.; Miao et al.). This project investigates the effectiveness of data imputation strategies on the prediction of energy consumption in a smart building. Using an open smart-building IoT sensor dataset, this study simulates sensor failure by introducing approximately 10%, 20%, 30% missing data under controlled conditions. Prior research demonstrates that model performance degrades progressively as missingness increases, making these percentages appropriate for evaluating both performance loss and recovery through imputation (Miao et al.; Casella et al.) Comparative evaluation is conducted to measure performance degradation due to missing data, the recovery of predictive accuracy after imputation, and the extent to which imputation enhances model robustness. The findings provide insight into the effectiveness of each data imputation technique for maintaining reliable machine learning performance in real-time building sensor systems under failure conditions (Miao et al.; Casella et al.).
Publication Date
Spring 4-9-2026
Recommended Citation
Lwanga, L.,Rungkittikhun, C. & Nyagesuka, D. (2026, April 9). A comparative study of imputation strategies for predicting energy consumption in a smart building with sparse IOT sensor data [Poster presentation]. Student Research Conference Spring 2026, Saint Paul, MN, United States. https://metroworks.metrostate.edu/student-scholarship/35
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Comments
Spring 2026: Student Research Conference