Multiple Self Organising Maps (mSOMs) for simultaneous classification and prediction: Illustrated by spoilage in apples using volatile organic profiles

• Main Authors: Siong-Fong, Sim, Sági-Kiss, Virág
• Format: E-Article
• Language: English
• Published: Elsevier B.V. 2011
• Subjects: GE Environmental Sciences
• Online Access: http://ir.unimas.my/id/eprint/15789/1/Multiple%20Self%20Organising%20Maps%20%28mSOMs%29%20for%20simultaneous%20classification%20%28abstract%29.pdf; http://ir.unimas.my/id/eprint/15789/; http://www.sciencedirect.com/science/article/pii/S0169743911001572

Self Organising Maps (SOMs) have been modified and extended for various applications. This paper employed multiple SOMs (mSOMs) in supervised manner for multitasking involving simultaneous classification and prediction providing more information on a sample. This was demonstrated on the GC–MS dataset of apple spoilage where the volatile organic compounds (VOCs) of two groups of apples, control and inoculated, were monitored over 2 to 10 days. Multiple SOMs were used to determine whether a sample is a control or an inoculated apple and at the same time the spoilage day is predicted, i.e. how many days an apple has been left with or without inoculation. Multiple SOMs are different from traditional supervised SOMs; in mSOMs, samples are divided into classes for training on different maps. This approach of SOMs does not require optimisation of the scaling value however, it is important to make sure that the clustering qualities of mSOMs are comparable. Growing Self Organising Maps (GSOMs) was incorporated to automatically determine the suitable map size with comparable clustering qualities based on the mean quantization error. The findings demonstrated that mSOMs can be potentially applied for simultaneous analysis allowing more information to be retrieved on a sample reducing the overall computational time of an analysis.