Soft Sensing as Class-Imbalance Binary Classification – A Lattice Machine Approach

• Published in: Ubiquitous Computing and Ambient Intelligence. Personalisation and User Adapted Services pp. 540 - 547
• Main Authors: Wan, Huan, Wang, Hui, Guo, Gongde, Lin, Song
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 2014
• Series: Lecture Notes in Computer Science
• Subjects: binary classification; class imbalance learning; Lattice Machine; Soft Sensing
• ISBN: 9783319131016; 331913101X
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-13102-3_85

Soft sensing is a class of problems that aim to sense something of interest that cannot be measured directly through something else that can be measured directly. The problems are usually studied as separate topics in different fields, and there is little research studying these problems in a unified fashion. In this paper we argue that there are commonalities among these problems. They can all be formulated as class-imbalanced binary classification problems. We present an extension of Lattice Machine, which is binary classification and by focusing on characterising positive class to deal with class-imbalanced binary classification problems. We also present experimental results, where some public data sets from UCI data repository are turned into binary-class data and consequently they become class-imbalanced. These experiments show that the extended Lattice Machine outperforms the popular machine learning algorithms (SVM, NN, decision tree induction) when used as soft sensing engines, in terms of precision.