Study of Stochastic Approach Impact on Accuracy of Sequence Element Classification Task
Author:
- 9 December 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Abstract
In some tasks, it is necessary to represent a process as a sequence of stages that should be classified. It is important to get correct stage classes because it impacts the performance of the entire task solution. In this paper, I consider the problem of low classification accuracy in those cases when stages of different classes may be similar to each other. I consider the solution to the problem in the example of the abnormal human action recognition task. In the solution, an action as is presented a sequence of discrete states where every state is described by a human pose. Also, the action process is considered as a stochastic process. In this paper, I propose an approach, the main new idea of which is that stochastic data are used in addition to image data describing a human action for recognition. In this approach, a stochastic graph of discrete stages is used as auxiliary data to describe the action process. The graph is parameterized by stage transition probabilities. These probabilities are computed explicitly from the training data. The proposed approach was tested on synthetic data. As a baseline, I use the performance of the classifier trained on the visual data only. The study shows that the approach doesn't outperform the baseline in most cases. This shows that the way stochastic data is used in this work is inefficient. The application of this approach to solving real-life problems requires additional research. The results of the work will be used to find other ways to use stochastic information to improve human abnormal actions recognition.Keywords
Funding Information
- RFBR (19-37-51028)
This publication has 16 references indexed in Scilit:
- Improvement of Bagging performance for classification of imbalanced datasets using evolutionary multi-objective optimizationEngineering Applications of Artificial Intelligence, 2020
- Assessment of Data Augmentation Strategies Toward Performance Improvement of Abnormality Classification in Chest Radiographs2019
- Deep Learning for Two-Step Classification of Malignant Pigmented Skin Lesions2018
- Classification Performance Improvement Using Random Subset Feature Selection Algorithm for Data MiningBig Data Research, 2018
- Data augmentation for improving deep learning in image classification problem2018
- Automatic two-channel sleep staging using a predictor–corrector methodPhysiological Measurement, 2018
- A new efficient training strategy for deep neural networks by hybridization of artificial bee colony and limited–memory BFGS optimization algorithmsNeurocomputing, 2017
- One-step and Two-step Classification for Abusive Language Detection on Twitter2017
- Machine learning based multi-physical-model blending for enhancing renewable energy forecast - improvement via situation dependent error correction2015
- Human action recognition by learning bases of action attributes and parts2011