MWITONDI, Kassim and YOUSIF, Adil (2012). A sequential data mining method for modelling solar magnetic cycles. Lecture Notes in Computer Science. (Submitted)Full text not available from this repository.
Solar magnetic activity has been studied by scientists for generations and to this day modelling its behaviour still poses a major challenge to the scientific data community. However, enhancements in methods of data acquisition and analyses provide hopes for increased predictive accuracy and reliability in the overall modelling of natural phenomena. Conventionally, modelling of natural phenomena has relied on the deployment of mathematical models - typically built on specific underlying assumptions which may be violated causing the models to fail to yield closed form or unique solutions. We propose an adaptive approach to modelling solar magnetic activity cycles using integrated adaptive unsupervised and supervised models. The approach involves two main steps - detecting naturally arising patterns in historical sunspots data using an adaptive clustering technique, labelling the emerging structures and applying a supervised model to perform predictions. Monthly sunspot numbers spanning over hundreds of years – from the mid-18th century to the first quatre of 2012 - obtained from the Royal Greenwich Observatory provide a reliable source of training and validation sets. Model selection is based on ROC assessment. Finally, we demonstrate how the method can be adapted to other applications involving natural phenomena of atmospheric, oceaonographic, meteorological or geological nature. Key Words: Clustering, Data Mining, Predictive Modelling, Solar Magnetic Activity, Sunspots, Supervised Modelling, Support Vector Machines, Unsupervised Modelling.
|Research Institute, Centre or Group:||Cultural Communication and Computing Research Institute > Communication and Computing Research Centre|
|Depositing User:||Kassim Mwitondi|
|Date Deposited:||24 Sep 2012 16:19|
|Last Modified:||24 Sep 2012 16:19|
Actions (login required)
Downloads per month over past year