I have tried to learn the ensemble ML algorithms like Bagging and Boosting. But I not clear about that why in case of Bagging algorithm we have to consider Decision trees parallelly as base model and in case of Boosting have to take Decision tree Sequentially.
Is there ant specific reason behind this? someone pls explain it.
Bagging is just an ensemble of classifiers, which all contribute to final decision. You can create ensemble using different features from your data (random forest), you can also train different models on same set of features.
In vanilla ML every record from data set is treated with the same weight. Idea behind boosting (like adaboost) is to train models iteratively and check with which records there're any problems. You're modifing weights accordingly, train next model and hope it'll do better. The idea from real world is: some records are easy, some are tough, so we're trying to train a model, which will be able to tackle both.
This is just intuitive look. There's quite few methods. Best to check docs of particular method, like xgboost.
It's also good to run them yourself on different data sets to acquire some intuitions like: vanilla SVM will fail on data with outliers, xgb will do just fine.
Ensemble learning is a technique that creates multiple models and then combine them to produce improved results.
And sometimes, if we have just 1 model which is trained from our dataset, that model might get overfit or it might suffer from a high variance problem. So, we can use ensemble learning to tackle that problem.
