Improving the accuracy of a tabular_learner model

Hello, I’m wondering if anyone can provide some advice on steps that I could take to improve the performance of a tabular_learner model. I have a classification task, with a data set of around 1.2m rows. The model is currently predicting at around 85% accuracy, I have tried a few different techniques to improve performance, but can’t increase the accuracy;

  • experimented with different learning rates
  • experimented with different numbers of layers
  • more / fewer categorical and continuous variables

Are there any other suggestions of steps that I could take to improve accuracy?

Thanks
Paul

Have you tried to experiment with regularization parameters (ps, emb_drop and etc)?