High Number of Machine Learning Algorithms Will Cause You to Miss the Point

If not approached wisely.

Photo by Ben White on Unsplash

Support vector machine, logistic regression, k-nearest neighbors, k-means clustering, random forests, XGboost, and the list goes on.

At the beginning of my data science journey, I thought I had to learn all the machine learning algorithms in detail. I wanted to do so as well because they all seemed intriguing and got my attention.

So, I spent hours studying. I learned most of them in detail. I admit it was fun learning them.

However, in most cases, the algorithm is only a small part of the data science workflow.

Don’t get me wrong. I’m not saying that what a machine learning algorithm does is not important. In fact, it's where the fun begins.

But, a substantial amount of time and work is done prior to using a machine learning algorithm. If you achieve to prepare the data correctly, you are at least 90% done.

At that point, the algorithm you choose can only make a small difference. This small difference is usually not worth spending hours unless you are in Kaggle competition.

In real life, 0.02 percent difference is not significant whereas it might help you jump over hundreds of teams in a Kaggle competition.

So, what do we do? We pick Xgboost and be happy :)

Putting the jokes aside, XGboost, LightGBM, CatBoost, or even a simpler model depending on the task will be just fine.

What is the point?

Photo by Mark Fletcher-Brown on Unsplash

The rich selection of machine learning algorithms will cause you to miss the point. But, what is the point?

The point is feature engineering.

Features are what define an observation (or data point). If we are trying to predict used car prices, age and mileage are two must-have features.

Whatever algorithm you choose will be useless without these features.