I am new to machine learning and I am having trouble with fitting a data set for a classification model. What I would like to know is after pre processing data and fitting to a model with just default hyper parameters, how much performance can I expect?
To clarify,
From the basic logistic regression model that I have trained, I am getting around 50% accuracy for both test and train sets, can I expect a big jump in performance with just hyperparameter tuning?
It may or may not.
The "may not" part :
If your cleaning, feature engineering and pre-processing steps are too shallow, you won't get a big increase in performance since you'll still be in a "garbage in, garbage out" step. You might also not have enough data, and your model is underfitting. In that case, hyperparameter tuning won't help either.
The "may" part :
I have used hyperparameter tuning tools like HyperOpt that significantly increased my model's performance. Once you start working on neural networks, tuning your learning rate will have a massive impact as well.
