Nominal attribute-based continuous type feature construction method

Abstract

The invention discloses a nominal attribute-based continuous type feature construction method. The method comprises the following steps of: 1) performing data preprocessing; 2) setting a feature construction frame according to business background knowledge; 3) generating concrete a feature construction path; 4) constructing corresponding features according to the feature construction path and generating a training set; (5) performing feature selection on the training set and constructing a prediction model; (6) saving the relevant data set and the prediction model and terminating an off-line training process; 7) performing preprocessing and feature extraction on sample data required to be subjected to on-line prediction; and 8) using a prediction model obtained through the off-line training to predict a sample. The nominal attribute-based continuous type feature construction method of the invention cannot only be applied to a user-item scene and but also be applied to more general classification and regression prediction problems with nominal attributes or categorical variable features. Compared with traditional One-Hot and Dummy coding, the features generated by using the method of the invention make the differences of samples more obvious and have strong interpretability.

Description

technical field [0001] The invention relates to the field of feature engineering in machine learning, in particular to a continuous feature construction method based on nominal attributes. Background technique [0002] With the advent of the era of big data and the rise of the Internet, various machine learning algorithms are used to mine the commercially valuable information contained in the data, and feature engineering is a key step in the machine learning system, which determines the performance of the system. The upper limit of accuracy, feature construction is an important part of feature engineering. At present, most of the feature construction is manual extraction based on rules, which largely depends on the engineer's understanding of the business background. It is difficult to extract comprehensive features in a short period of time, especially for nominal attribute features or categorical variable features. Color features such as "yellow, red, and blue" often con...

AI Technical Summary

Problems solved by technology

At present, most of the feature construction is manual extraction based on rules, which largely depends on the engineer's understanding of the business background. It is difficult to extract comprehensive features in a short period of time, especially for nominal attribute features or categorical variable features. For color features such as "yellow, red, and blue", the nominal attributes are often converted into sparse vectors with the same length between each other, and the features are constructed by One-Hot encoding or Dummy encoding.

Although each dimension of this coding method represents whether a certain nominal attribute or categorical variable appears, it has a certain physical meaning, but for different samples, this feature representation defaults to the same fixed value as the distance between each other. One point may be contrary to reality, and when the nominal attribute has too many values, this encoding method will lead to the problem of excessive feature dimension

Images