Use of machine learning algorithms in monitoring environmental pollution

Increasing population, industrialization, disposal of toxic wastes, and unhealthy agricultural activities increase soil pollution and disrupt the environmental balance. Hydrocarbon pollution is one of them. For both ecosystems to remain in their natural structure, and for human health not to be adversely affected, the boundaries of the hydrocarbon pollution that occurs must be determined, monitored, and intervened correctly. While conducting chemical experiments is one method, another method is to examine the electrical properties of the environment at certain time intervals. The geophysical spectral induction polarization (SIP) method, which can scan larger areas compared to chemical experiments and do this without damaging the samples taken, has become popular in recent years. Various metallic minerals - such as magnetite - increase in areas contaminated with hydrocarbons over time. The SIP method, which can detect the electrical properties of different types of metallic minerals, especially magnetite, is used to determine the boundaries of this type of environmental pollution. Laboratory experiments are needed to develop this method. However, these experiments can usually be time-consuming and costly. Machine learning algorithms that learn patterns and relationships formed with data have the potential to perform some operations more effectively, prevent loss of time, and minimize costs. In this thesis, electrical responses of magnetite, a metallic mineral, collected in a laboratory environment were used. SIP results were tested with different machine learning algorithms and evaluations were made about the performances of the algorithms. The performances of the machine learning algorithms used in the thesis were compared with real data, and 13 algorithms that showed over 95% success were determined in this process. It was determined that Random Committee and Random Tree algorithms had the lowest error rates and exhibited the best performance with fast working times. In addition, it should be taken into account that the success rate of other algorithms was 95% and above. This thesis provides important information on how machine learning algorithms can be used to create a scale for the SIP method (in laboratory conditions)used in environmental pollution analyses caused by hydrocarbons. Finally, it is recommended that machine learning algorithms, which have the potential to reduce the time and costs of laboratory studies, and help the obtained data to be free from human-induced errors, be used to create the scale to can be used for determining and monitoring environmental pollution limits.