This study aims to create new methods from problems that we will meet or face in ICIS 2018 Challgne that commen method was not the best option for these problems
Results are 2 methods one in segmentation called limited crop certain(LCC) and one in enseble call class weight transformation(CWT)
we did not aim to get high results we just want to prove that our methods worth to try as replacement for common methods
heightest results we could achive was by new enseble (CWT) (10 simple models no 5-fold) 83,4% balanced accurcy on live leader board without extra data
(LCC) was compared with crop certain by simple analysis and with bitwise(AND mask with image) by models
recommendation if you want to understand just the 2 method foucs reading only on start of explaination of each method in both chapters('start' pre-processing architectures for (LCC) ,'afte middle' train and results for(CWT)