Question-56: What do you mean by training set?
Answer: Whatever Machine Learning Model you create those were trained on some certain datasets. And that dataset is called training dataset. And during the model learning phase Model learn from the inherent patterns in the data and loss function or objective function is reduced through training. If the trained model works or perform same as while training it did with respect to performance then you can say your model is generalized as well.
Question-57: What is the main objective of the Machine Learning Model tuning?
Answer: The main point of Machine Learning Model tuning is to generalize it, so that it can perform better with the new data.
Question-58: Can you help in understanding of Machine Learning Model generalization in statistical term?
Answer: Generalization, means fitting your Machine Learning Model or algorithm on the sample data to eliminate the effect of Outliers, unstructured pattern, seasonality which exists in your data.
And having Generalized model do predations on the new data which is more concrete, and having less variance in it with the minimum bias-ness.
Hence, you should even apply the generalization on your training data set as well before actually you use them to train your model.
Question-59: Can you show using graph what is Generalization and overfitting?
Answer: As you can see in below image. Image in left represent a model which is over trained and had been overfitting. And would perform worst with the new real data. And you can also see this model is highly complex as well. But you see image in the right which is a generalized model and perform best when represented with the new data. Your target should always to create most generalize model, so that you can predict it very well.
Question-60: What are your expectations from Machine Learning Model?
Answer: We need Machine Learning Models to describe the underline pattern in the data. However, when you work with the real data. It does not come without the noise. And we want that model capture the signal in the data and not the noise.
