Abstract

Modeling techniques were used to study minerals in clay fraction of soils from Capricorn District, Limpopo Province, South Africa. Minerals in the clay fraction of soils were identified by X-ray diffraction (XRD) technique and semi-quantified. The minerals were then subjected to a combination of exploratory data analysis (EDA), graphicalvisualisation and domain-partitioning techniques in order to determine their cross-influence to one another in terms of abundances. Quartz and kaolinite were major dominant minerals in the soils; smectite, feldspar and mica were in minor to trace quantities. Consensual associations among other traces and high quantities of minerals were detected. Evidence of relationship using EDA portrayed general skewness in favour of quartz and kaolinite. Quartz remained dominant in the soils but with a consistent high probability of co-existence with kaolinite. Where there is low quartz content, kaolinite increased with the drop in quartz made up for by a combination of smectite, mica and feldspar. The nested nature of interaction also revealed indirect relationship between quartz and mica. The tree model, which yielded 100% accuracy, showed smectite as the first important mineral in identifying whether there is high, medium or low quartz content in the sols. Down the line the model relies heavily on both mica and kaolinite. Collating the minerals contents and data modeling procedures, inter alia, it could be inferred that the weathering of feldspar and mica may have an impact on the mineralisation of kaolinite and smectite; which are both important minerals in several agricultural applications.

Key words: Conditional probability, decision trees, domain-partitioning, feldspar, kaolinite, smectite.