Looks like the contra community is liking stuff on machine learning, and I thought it would be interesting to share another article we put together about why machine learning models can underperform for... just no obvious reason.

Despite having clean data and tuned hyperparameters, the accuracy graph barely moves. The trick here is everything related to feature engineering. We wrote a piece titled “Feature engineering decides machine learning outcomes” to share our thoughts on this quasi-"mystic" exercise.

I call it quasi-"mystic" because it is part art and part science. You have to be able to see through the rubble of messy data and know what will really move the needle for you. Much like the jeweller in our cover. It breaks down the essentials, from understanding your data’s structure to choosing well-validated features. Read the full piece here: https://www.algorithmic.co/blogs/feature-engineering-decides-machine-learning-outcomes