Full Description

In this paper, we present a comparative study of five methodsfor the estimation of missing values in building sensordata. The methods that were implemented and evaluatedinclude linear regression, weighted K-nearest neighbors(kNN), support vector machines (SVM), mean imputationand replacing missing entries with zero. Usingdata collected from an actual office building, the methodswere evaluated using varying parameter settings. Correlationbased feature selection is used to evaluate how usingdifferent subsets of attributes may affect each methodsperformance. We also evaluate the effect of includinglagged variables as predictors. To test the robustness ofeach method, the amount of missing values were variedbetween 5% and 20%.

Citation: ASHRAE/IBPSA-USA Bldg Simulation Conf, 2016

Product Details

Published: 2016 Number of Pages: 8Units of Measure:DualFile Size: 1 file , 1 MB Product Code(s): D-BSC16-54