2018-08-11 Titanic的kenel

Introduction to Decision Trees (Titanic dataset)

That's why advanced techniques such as Deep Learning or Ensemble Learning (cf. Anisotropic Kernel) are commonly used for complex tasks. But remember the KISS principle (Keep It Simple, Stupid)! Always consider the complexity/accuracy trade-off: complex techniques should only be used if they offer significant improvements. Simpler models are also less prone to over-fitting and tend to generalise better.

But if we're using Machine Learning to actually get insights from the data, "blackbox" models are almost useless and it's best to stick with simpler, transparent techniques. Let's take the case of a supermarket looking to better understand customer behaviour: the straightforward Apriori algorithm can quickly offer relevant insights like "80% of customers who bought a suit also bought a tie" so they may try to increase tie sales by offering a discount to clients buying a suit . Of course, a complex classification algorithm will do better at identifying the customers who bought a tie by taking into account more features, but is that really useful for the supermarket?