Automated Feature Engineering
Data Mining, Classification
Feature Construction, Feature Selection, Representation Learning, Numeric Features
Data Scientists often find a central step in their work, is to implement an appropriate transformation restructuring the originally given data into a new and more revealing form. For areas where large amounts of training data or intensive computational sources are not available, feature engineering is a manual effort, needless to say, also tedious and non-scalable. This technologypresents an approach to learn a generic representation by mining pairwise feature associations, identifying the linear or non-linear relationship between each pair, applying regression and selecting those relationships that are stable and improve the prediction performance.
Type of Work
Current State of work
Technology designed and implemented
1. Can be used for mining datasets with numeric attributes.
2. Specially useful for biomedical/gene expression type of data, that is characterized by high dimensionality and lower sample space.
1. AutoLearn - Automated Feature Generation and Selection. ICDM, IEEE 2017, pp. 217-226
2. Data Driven Feature Learning, ICML 2017.