I'm classifying traffic using the CIC-2020 Darknet data set. There are 8 classes in the data set. I tried a lot of different ways. (Over sampling, Under Sampling, Complex Sampling and Outlier Elimination, Feature Selection, model change - cat boost, rus boost, Decision Tree, Random Forest, OvO, OvR, SVM etc..) However, the performance doesn't improve. In the picture below, I only selected features using RFECV and used Light GBM. This has the best performance.

LightGBM Performance

Is there a good way to improve performance in multi-class classification? Please help me.