Deep Learning Will Make Classical ML Go Extinct?

“The future depends on some graduate student who is deeply suspicious of everything I have said.” — Geoffrey Hinton(Godfather of Deep Learning)

Deep learning is starting to rule the minds of data scientists over the last decade. The modern hardware is so strong and capable that we can fit and train even terabytes of data on a neural network. Now, beginner data science guys ask a smart question, “Why should I learn the classical linear regression, logistic regression, random forest, K-NN, etc.?” Should you skip the classical machine learning algorithms and directly go for the big thing?

The answer is no! Classical ML algorithms are not even close to even getting the name ‘endangered’. Their importance and use will vary and likely be continued by corporates. Let us dive deep into its justification…

1. Deep Learning is Not Always Stronger than ML algorithms

Classical ML algorithms outclass deep learning for small datasets

Deep learning was created while keeping in mind that it will be fetched a large size of data. The very algorithm of deep learning works on millions of trials and errors. This requires a humongous size of data. Now, it’s not possible to always get this much large data.

So, in problems where the size of data is less, their data scientists merrily go for classification machine learning algorithms (mostly choosing random forest).

2. When the Dataset is average sized

If you receive data that is neither big nor small, in that case deep learning might not be the best option all the time. In these kinds of dataset sizes classical machine learning algorithms perform practically the same as deep learning. So, you can opt for the classical ones. This is because they take much lesser time for model training.

3. Classical ML Serves as a base for Data Science Trainees

Frankly, a deep learning mechanism isn’t a child’s plaything! Though the implementation has become easy, in order to harness the full capabilities, you still need to learn its proper functioning.

Due to this, it is better to start your Data Science journey with Classical ML algorithms and practice their projects for the first few months. This will give a good exposure to data preprocessing and other common stuff that you have in both deep learning and classical algorithms.

This is a smart way to learn deep learning.

Thank you