Wearable device activity prediction method based on federated personalized random forest

Abstract

The invention discloses a wearable device activity prediction method based on a federated personalized random forest. Each user generates a respective hash table through employing a local sensitive hash function, and each user finds a similar user through comparing the number of similar sample data between hash tables of all parties. Each user can initiate training actively, only users similar to the user are searched for joint training of a random decision tree, and the random decision tree is owned by the users participating in training. The privacy of a user is protected by using a differential privacy mechanism in the node division process of a random decision tree, each party disturbs the calculated information gain, and the disturbed information gain of each party is aggregated at one party to obtain the optimal division in a candidate attribute set. For the generated random decision tree, the users obtain their own final random forest models through incremental selection, the random forest models of the users are different from each other, and the personalized model is more suitable for the users to predict their own data.

Description

technical field [0001] The invention relates to the technical field of federated learning, in particular to a method for predicting activities of wearable devices based on federated personalized random forest. Background technique [0002] Activities of daily living are closely related to people's health. In recent years, with the development of wearable technology, people use smart phones, wristbands, smart glasses and other wearable devices to track their activities so as to understand their health status. Artificial intelligence technology will also be integrated into wearable devices to achieve smarter and more self-aware applications. This trend has already begun to emerge at the CES2014 Consumer Electronics Show. For example, according to the wearable device to collect user-generated activity data and a series of data such as the surrounding environment, use the trained machine learning model to accurately identify the user's ongoing activities, so that smart wearable...

AI Technical Summary

Problems solved by technology

[0003] What the present invention aims to solve is the problem that personal sensitive information is easily leaked in the process of implementing artificial intelligence in existing wearable devices, and provides a wearable device activity prediction method based on federated personalized random forest

Images