Çevresel enformatikte veri madenciliği ve makine öğrenmesi

Nowadays, environmental informatics is one of the fields where data mining and machine learning techniques for data processing & analysis are frequently used today. In this thesis, several case studies including environmental data on different subjects were conducted for different tasks (classification, regression, clustering, association rule mining, time series) under data mining and machine learning. Examining the data obtained from air quality monitoring stations and meteorological data, determining which station type (industrial, rural, or urban) a particular air quality monitoring station belongs to, and making the most accurate clustering application according to the similarities of the stations are a certain part of the experimental studies. In addition, the relations between air pollutants and meteorological factors were examined in detail within the scope of association rule mining analysis, and the relationship rules were drawn. On the other side, regression and time series analysis for soil temperature prediction at various depths were performed. The proposed methods to handle air and soil-related problems are as follows: (i) Enhanced Bagging (eBagging) which is a new ensemble learning method, (ii) Soil Temperature Prediction via Self-training (STST) which is a new semi-supervised learning method, (iii) Majority Voting Based Multi-Task Clustering (MV-MTC) which is proposed to consider multiple tasks jointly, (iv) WARM which is a new type of weighted association rule mining, and (v) Multi-View Multi-Depth Soil Temperature Prediction (MV-MD-STP) which is a new soil management framework. Considering the experimental results, it is clear that the newly proposed methods performed better than the other state-of-the-art methods they were compared to and offered effective solutions to environmental problems.