An application of data mining and knowledge discovery process in the field of natural gas exploration

This thesis analyzes the process of data mining and knowledge discovery (DM&KD) in the area of natural gas exploration. The process consists of five steps: Problem Definition, Collecting the Data, Data Pre-Processing, Application of the Main DM&KD Algorithms, and Interpretation of the Results of the DM&KD Process. The problem is to identify the DM & KD algorithm among a set of algorithms that extracts the highest number of useful valid rules targeted to find natural gas deposits in sandstone rock type of Osmancik formation in Degirmenkoy natural gas field of Turkey. The data is collected along five wells from the gas field. In the Pre- Processing phase, outliers and erroneous data are removed. Then these data is transformed into Well Log SQL database and necessary connections are made between the database and Weka Data Mining Tool. Non-Nested Generalized Exemplar (NNGE), Predictive Apriori (PA) and PART algorithm are applied using Weka's toolset. In the gaseous sandstone zones of all Wells, at total best results in terms of valid useful rule amount is attained by NNGE algorithm. PA algorithm presents better performance than PART algorithm in the analyzed zones. For nongaseous zones, same ranking is achieved as in the gaseous sandstone zones. In addition to these results, interesting findings related with some of the extracted rules' explanatory power are determined.