The estimation of maximum directed verosimilitude (TMLE) helps you explain patterns in which other techniques are not enough
Neuronal networks can detect patterns, correlations and trends with amazing precision. But when it comes to responding “why did this happen?” They are as clueless as a parrot that mimics human speech.
Surely they will give you predictions, but try to ask for an explanation and you will look at a black box.
This limitation is not exclusive to neural networks. Correlation -based methods, such as linear regression and even advanced tools such as propensity score matching, cannot reach the core of trends based on causality in complex data. This is a problem when those who make decisions (read: their managers) demand processable business information and non -geek statistics that only make nerds happy.
At the risk of contradicting me, here is a very geek topic for you: Estimation of maximum directed verosimilitude (TMLE). The point is that TMLE is the best of both worlds. It allows you to play with numbers as much as your nerd brain wants, but also makes your managers happy to produce business information.
Basically, you get the rigor of causal inference plus the flexibility of automatic learning. This…
