Embedded Files
In this step, we are reducing the number of features to simplify our later machine learning models, and shorten their training time.
In this step, we are reducing the number of features to simplify our later machine learning models, and shorten their training time.
We will be applying two types of feature extraction:
We will be applying two types of feature extraction:
Feature Selection - selects columns/features in our data that are most relevant and helpful to our predictive problem
Dimensionality Reduction - make new combinations of features based on a lower representation of the data (i.e. reducing the dimensions of our feature set)
Methods
Methods
1. Random Forest
1. Random Forest
Random forests are made of decision trees, where the nodes are conditions on features that are designed to split the data set so that similar response values end up in the same set (i.e. the same side of the decision tree).
Random forests are able to calculate feature importance by computing how each feature decreases the impurity of the tree.
Random Forest would be an example of a feature selection method.
2. Principal Component Analysis (PCA)
2. Principal Component Analysis (PCA)
Principal Component Analysis (PCA) maps a high-dimensional set of data to a lower-dimensional space, taking a large number of possibly correlated data and returning a set of linearly uncorrelated variables (principle components).
PCA is an example of a dimensionality reduction method.
3. Auto-encoder
3. Auto-encoder
Auto-encoder is a neural network (set of algorithms that recognize patterns, modeled after the human brain) that copies its input to its output by learning to represent (encode) a set of data in a lower dimension then decoding it to return the data in its original dimension.
Auto-encoder is the second method of dimensionality reduction that we used in this project.
Page updated
Google Sites
Report abuse