In machine learning algorithms, one of the biggest problems is
scarcity of time and requirement of huge datasets. Feature selection
provides optimization for these matters. Reducing time and amount of
data required can change the fitness of the model significantly. By aiming
this, feature selection algorithms are very important.
In order to see those effects, we consider implementing the
algorithms which are “consistency-based” and “ correlation-based”
feature selection. In addition to that, we can also compare the required
amount of time and data to reach the same fitness levels for the model.
To pick up a model, we have had a lot of different options but
among them, we have decided on Bank Marketing Dataset. One of the
most desired features of our project was finding an appropriate data. We
have searched on the internet and it has two essential properties that
first, high number of features and second, a lot of instances available.
In the model, the classification goal is to predict if the client will
subscribe (yes/no) a term deposit.
Overall, we would like to reduce number of attributes and to
decrease least amount of time to train the model for the same fitness
levels.
- Homework #1 --> Perceptron Learning Algorithm, Linear Regression, Nonlinear Transformation implementations from scratch.
- Homework #2 --> Gradient Descent, Logistic Regression, Regularization with Weight Decay implementations from scratch.
- Homework #3 --> SVM training & testing and cross-validation in LIBSVM.