I've been working on a project where it consists of a dataset of profiles (50k rows) along with their position, age group and hobbies (200 columns). These features (except for the position) are graded from 0 to 10. A value near 0 means that they have no interest in that particular hobby or they are not part of the age group and a value near 10 means that they are more likely to be into this hobby(or in the age group etc...)
Some of the profiles are not graded for some hobbies. So, I decided to put a 0 so as they are not into this hobby. I have also deleted the columns that had more than 80% of empty values.
The profiles have 2 binary categories. I want to distinguish these categories with a training model. So I applied PCA to diminue the dimensions and then Logistic Regression for the classification. I got an accuracy of 0.68. So I was wondering if I may have preprocessed the data wrongly or could the model be improved in some way that I missed ? I also tried DecisionTreeClassifier and tried to prune the decision tree but still the same accuracy so that made me think maybe it comes from earlier steps?