Preparing data for machine learning
View Patent ↗An apparatus and methods for feature selection and classifier builder are disclosed. The feature selection apparatus allows for removal of bias features. The classifier builder apparatus allows building a classifier using non-biased features. The feature selection methods disclosed teach how to remove bias features. The classifier builder methods disclosed teach how to build a classifier with non-biased features.
1. A method of machine learning, comprising:
obtaining input training data that include a plurality of data items, individual data items within the data set including a first label, a second label and a feature vector, the feature vector specifying values for a plurality of features;
assigning a first predictiveness value to features within the plurality of features based on the first label;
assigning a second predictiveness value to features within the plurality of features based on the second label;
generating a third predictiveness value for features within the plurality of features based on said first predictiveness value and said second predictiveness value;
providing output training data that include the feature vectors, the second labels and the third predictiveness values; and
generating a classifier by performing a machine learning process on the output training data.
2. The method of claim 1 further comprising:
using a threshold value, to determine a number of features with said third predictiveness values that are input to the machine learning process.
3. The method of claim 1 further comprising:
using a threshold value, wherein only features whose values of said third predictiveness value exceed equal or exceed the threshold value are input to the machine learning process.
4. The method of claim 1 , wherein the first label for a given data item indicates a source of said given data item.
5. The method of claim 1 , wherein the second label for a given data item comprises knowledge about a class to which said data item belongs.
6. The method of claim 1 , wherein the second label indicates whether a particular data item is e-mail spam.
7. A computer-readable medium storing computer-executable process steps for machine learning, said process steps comprising:
obtaining input training data that include a plurality of data items, individual data items within the data set including a first label, a second label and a feature vector, the feature vector specifying values for a plurality of features;
assigning a first predictiveness value to features within the plurality of features based on the first label;
assigning a second predictiveness value to features within the plurality of features based on the second label;
generating a third predictiveness value for features within the plurality of features based on said first predictiveness value and said second predictiveness value;
providing output training data that include the feature vectors, the second labels and the third predictiveness values; and
generating a classifier by performing a machine learning process on the output training data.
8. The computer-readable medium of claim 7 , wherein the first label for a given data item indicates a source of said given data item.
9. The computer-readable medium of claim 7 , said process steps further comprising:
using a threshold value, to determine a number of features with said third predictiveness values that are input to the machine learning process.
10. The computer-readable medium of claim 7 , said process steps further comprising:
using a threshold value, wherein only features whose values of said third predictiveness value exceed equal or exceed the threshold value are input to the machine learning process.
11. The computer-readable medium of claim 7 , wherein the second label for a given data item comprises knowledge about a class to which said data item belongs.
12. The computer-readable medium of claim 7 , wherein the second label indicates whether a particular data item is e-mail spam.