Machine learning method based on weighted depth forest

Abstract

The invention provides a machine learning method based on weighted depth forests, wherein the depth forests include multi-granularity scanning and cascading forests; multi-granularity scanning can generate corresponding class vectors by acquiring a plurality of feature subsets, and the generated class vectors are spliced into the original sample feature space as the input features of the followingcascaded forests; cascaded forests are used to realize token learning and include random forests and completely random tree forests. Forest cascades are formed by hierarchical structure. By calculating the corresponding weight of prediction accuracy of each sub-tree in each forest, and then summing up the prediction probability vectors of each sub-tree by weight, the prediction results of the whole forest can be found out. It not only improves the prediction accuracy of depth forest, but also reduces the cascade series.

Description

technical field [0001] The invention relates to a weighted deep forest machine learning method, which is especially suitable for application fields such as image processing and audio analysis. Background technique [0002] Both Deep Forest (DF) and deep neural network use multi-level structure for representation learning, but deep forest makes up for the shortcomings of deep neural network with its simple training model and the characteristics of not relying on a large amount of data for training, and It is gradually applied in engineering practice. Reference 1: Zhou Z H, Feng J. Deep Forest: Towards An Alternative to Deep Neural Networks [J].arXiv preprint arXiv:1702.08835.2017. [0003] Deep Forest consists of two parts: Multi-Grained Scanning and Cascade Forest. Among them, multi-granularity scanning obtains multiple feature subsets through sliding window technology to enhance the diversity of cascaded forests. Cascade forest is a forest composed of decision trees to a...

AI Technical Summary

Problems solved by technology

Since the prediction accuracy of each subtree in the forest is different, the arithmetic mean will cause the wrong prediction of the subtree to affect the prediction of the whole forest, and as the number of stages increases, the error may be further amplified

Images