5Model Reduction for Support Vector Machines
5.1. Introduction
In Chapter 1, we stated that there are a great number of factors which probably impact energy dynamics of a building. In Chapter 4, when predicting one single building’s energy profiles, we selected 24 features to train the model, including variables from weather conditions, energy profiles of each sublevel component, ventilation, water temperature, etc. While predicting multiple buildings’ consumption, four more features which represent building structure characteristics were used. However, we do not guarantee that these features are the right choices, nor can we even say they are all useful. How to reasonably choose a subset of appropriate features to be used in model learning is one of the key issues in machine learning. On the one hand, using different sets of features would probably change the performance of the models in accuracy and learning speed. On the other hand, the optimal set of features would make the predictive models more practical.
In this chapter, we discuss how to select subsets of features for SVR applied to the prediction of building energy consumption [ZHA 12a]. We present a heuristic approach for selecting subsets of features in this chapter, and systematically analyze how it will influence the model performance. The motivation is to develop a feature set that is simple enough and can be recorded easily in practice. The models are trained by SVR with different kernel methods based on three ...
Get Data Mining and Machine Learning in Building Energy Analysis now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.