In this dataset, there are 33 variables. The 33rd column (G3) is the output column, or dependent column, which varies from 1 to 20. The other 32 columns are input variables or (independent variables). We use two types of techniques for taking in the input of 32 columns, and then predicting the 33rd column based on that.
The two techniques we have used are: Support Vector Regression (SVR) Multi-perceptron feedforward Neural Network (NN)
We divide the dataset into ‘training’ and ‘testing’. The dataset consists of two student sets, namely, Portuguese and Mathematics. We operate on both of them separately. We apply the fitting and calculate the Root Mean Square Error (RMSE) for both the techniques and both the datasets.
The SVR has performed better than NN in this case. The results are displayed in the Python Notebook itself.
DATA PREPROCESSING We apply Data cleaning based on the values they have, namely, numeric, ordinal and nominal. As string values are not very suitable to be directly passed to training networks for analysis.
The nominal data consists of non-numeric data, without any natural orderding, such as names of people, “Rajan”, “Krishna”, etc. These are each assigned a separate number, and then passed into the network, so that the model can fit it mathematically. Similarly, ordinal is where the data is non-numeric although the natural ordering may exist. E.g. Weight of a person, Low, Medium, High
Although they are non-numeric, there is still a natural ordering of this. Low – 1, Medium – 2, High - 3