This repo created by writting model training and infern process for kaggle competition. I used there the:
XGBoost Classifier - the most better prediction score
CatBoost Classifier
Keras Multi-Head Architecture Model - the main idea is using the special model for each class in dataset with 'sigmoid' activation fn for model output.
Simple Keras Classification Model - Just use the 'softmax' in the output layer and SparseCategoricalCrossentopy as loss
Notebook on kaggle: https://www.kaggle.com/code/yannchikk/multiclass-prediction-train-infern/
Competition: https://www.kaggle.com/competitions/playground-series-s3e26/
Dataset: https://www.kaggle.com/competitions/playground-series-s3e26/data/