HPR2955: Machine Learning / Data Analysis Basics

In this episode, I talk about different techniques that we can use to predict the outcome of some question depending on input features.

The different techniques I will go through are the ZeroR and OneR that will create a baseline for the rest of the methods.

Next up, we have the Naive Bayes classifier that is simple but powerful for some applications.

Nearest neighbor and Decision trees are next up that requires more training but is very efficient when you infer results.

Multi-layer perceptron (MLP) is the first technique that is close to the ones we usually see in Machine Learning frameworks used today. But it is just a precursor to Convolutional Neural Network (CNN) because of the size requirements. MLPs have the same size for all the hidden layers, which makes it unfeasible for larger networks.

CNNs, on the other hand, uses subsampling that will shrink the layer maps to reduce the size of the network without reducing the accuracy of the predictions.

Some references for further reading on Wikipedia.

https://en.wikipedia.org/wiki/Naive_Bayes_classifier
https://en.wikipedia.org/wiki/Nearest_neighbor_search
https://en.wikipedia.org/wiki/Decision_tree
https://en.wikipedia.org/wiki/Support-vector_machine
https://en.wikipedia.org/wiki/Multilayer_perceptron
https://en.wikipedia.org/wiki/Convolutional_neural_network

A video I made some years ago where you can see some visual aids for this subject.

https://youtu.be/Xys1N_7MbSs