When training a model on a dataset, it's essential to ensure that the unseen data (test data) will have the same columns as the training data. In your case, the categorical variables in your training and test datasets have different unique values.

You need to decide on a method to resolve this mismatch. One approach is to add missing features to the test dataset, filling them with zeros for features that are present in the training dataset but absent in the test dataset, and vice versa.

IN CONTEXT OF ONE-HOT ENCODING

Make sure the columns you have selected in the train dataset have :

1. One same data type for the whole dataset .

2. number of columns in Train dataset is "n" columns and the number of columns in Test is "n-1" columns.