I have a large dataset D which I balanced using under sampling method called RandomUnderSampler from imblearn package which reduce the class data with majority. The data have three classes: Yes (1), No (0), Unfinished (2).
This is the minimal 3d projection of the dataset after perform PCA on the balanced version of the dataset:
I tried RandomForestClassifier,BaggingClassifier, KNN and MLP Classifier etc with some hyperparameter tuning even LocalOutlierFactor via sklearn package but I cannot get desired precision even if I am using balanced data.
I am sharing the HTML version of the code I have so far in the following link : https://imbalance-data-classification.netlify.app/
All I see from google search is to balance and try different classifier while tweaking the hyperparameter which is not working. I atleast need precision of 80% rather than existing 44% I got in some instance of MLPClassifier.
What can be done given this situation to improve the precision for predicting data which should be labelled as Yes or No. My knowledge related to Machine and Deep Learning is limited.
