I have a binary classification problem. I have a somewhat good baseline (logistic regression) and I use lightgbm in a relatively standard fashion for prediction and explaination (setting contrib = True when predicting). I was thinking about using the relatively explainable baseline in the lgbm model as the baseline is generally used as reference.
One option seems to use the baseline score as a feature. Another option seems to be to use it as initial score. The baseline score as feature seems quite straigforward and play nice with prediction and explanation.
But using the baseline score as initial score option: how does it play with prediction and explanation steps exactly ? More generally what would be the advantage of using the baseline as initial score over the feature approach ?