Makine öğrenmesi ve CBS yardımıyla sıvılaşma değerlendirmesi ve haritalamasına yönelik Balıkesir'de bir vaka analizi

Liquefaction, which is defined as the loss of bearing capacity in liquefiable layers temporarily, is a soil problem that creates serious results in terms of soil - structure interaction. Although this problem can be solved owing to various soil improvement methods, firstly the liquefaction analysis must be determined correctly in the region where the structure is planned to be built. Therefore, in this thesis, the possibility of using artificial neural networks, random forest, and support vector machine, which are three well-known machine learning models, to model the complex relationship between liquefaction risk and soil seismic features was investigated to assess whether a soil is liquefiable or not. Liquefaction susceptibility mapping is critical for a study area located in high-risk earthquake zones. Hence, the susceptibility against liquefaction for the random study area in Balıkesir bay was investigated in terms of geologic and geotechnical aspect according to 7 actual borehole values were taken from near the study area. A determination of liquefaction status is vital for this region since the area is located on alluvial soil with high level of ground water table. The liquefaction potential (FS) of the study area was determined according to the SPT-based simplified method as well as the liquefaction potential index (LPI) was calculated and related micro zonation maps were created thanks to the geographical information systems. The obtained results show that the entire study area has a moderate level of liquefaction potential during a possible earthquake practically. Keywords: Liquefaction potential, Machine learning, Random Forest, Geographic information systems, Standard penetration test.