In this project, we seek to implement the ordinal classification paper here: http://www.cs.waikato.ac.nz/~eibe/pubs/ordinal_tech_report.pdf. A regular classification would treat each class as not having any order and would miss the inherent information that comes with ordering and this paper suggests a method to overcome this disadvantage.
Python has a package called mord which performs multiclass ordinal classification.
The general idea is to build k - 1 classification models each predicting the probability of a data point being greater than a class value.
We create a 1/0 field for all but final class and assign the value 1 if the class label is greater than the class of current column. We then go on to build classification models that predict the probabilty of 1/0. Once we have the probabilities, we derive the probabilities of current data point as belonging to each class. The class which has highest probability is assigned to the data point.
We use the winetesting dataset from UCI Machine learning repository: https://archive.ics.uci.edu/ml/datasets/Wine+Quality
Proceed here for the analysis