A regression algorithm for rock magnetic data mining

Document Type

Journal Article


World Scientific and Engineering Academy and Society (W S E A S)


Computing, Health and Science


School of Computer and Information Science




Originally published as: Guo, W. (2005). A regression algorithm for rock magnetic data mining. WSEAS Transactions on Information Science and Applications, 2(6), 671-678. Original article available here


The correlation between magnetic susceptibility and the content of magnetite is important in interpretation of magnetic anomalies, and understanding the basic magnetic behaviour of rocks in rock magnetism study. A few correlations were proposed in the 1950s-1960s and have been widely used. In the last a few decades, the adoption of new technologies in chemical analysis and magnetic measurements have led to the acquisition of more new rock magnetic data in a format different from the data obtained before. There is a need to establish new correlations between the susceptibility and magnetite content in rocks using the collection of data from both current and previous studies. The statistical analysis used in the previous studies was predominantly focused on seeking a sole linear or power correlation between susceptibility and magnetite content. This is because each study used the data collected from a confined area where the magnetite content in rocks was naturally concentrated within a certain range. Multiple correlations may be determined for different ranges of magnetite content if a database consisting of the data from various regions is used for the statistical analysis. In this study, data mining technique is introduced to rock magnetic data analysis. A data mining algorithm is designed to carry out data selection, pre-processing, transformation, and data analysis for rock magnetic data. This algorithm is able to search for linear, power, logarithmic, and exponential correlations that may exist between susceptibility and magnetite content contained in rocks. This algorithm is tested using a new magnetic database constructed by collecting the datasets from the previous studies. The results from this data mining process are then interpreted incorporating with the relevant knowledge in rock magnetism. Although strong linear, power, logarithmic, and exponential correlations are all revealed by this algorithm, only the power and exponential correlations are considered useful for different ranges of magnetite content in rock magnetism study.