SOFTWARE COMPLEX FOR FORECASTING MIXED TIME SERIES USING CLUSTERING
Authors:
1.
Kazan National Research Technical University named after A.N. Tupolev
Type:
Article
Pages:
from 103
to 110
Status:
In work
Language:
Russian
Keywords:
TIME SERIES FORECASTING, SOFTWARE PACKAGE, CLUSTERING, DBSCAN, LSTM, DATA PREPROCESSING, TIME SERIES PARAMETERIZATION, MACHINE LEARNING, ASP.NET CORE, PYTHON
Abstract (English):
The article presents a software package for time series forecasting based on clustering and deep learning methods. The package includes a control controller, modules for data preprocessing, time series parameterization, clustering using the DBSCAN algorithm, and forecasting using LSTM recurrent neural networks. An important feature is the parameterization of time series before clustering, which allows you to highlight key statistical and spectral characteristics of the data, such as mean, variance, ARIMA coefficients, trend, number of peaks and troughs. The use of the DBSCAN clustering algorithm provides automatic formation of clusters without the need to specify their number in advance, which makes the algorithm more flexible and resistant to noise. The use of separate LSTM models for each cluster allows you to take into account specific data patterns within a group, increasing the accuracy of predictions. The implementation is made in Python and ASP.NET Core, which simplifies integration into various information systems and ensures cross-platform. The article discusses in detail the architecture of the software package, its functionality, data processing algorithms and forecasting methodology. Experiments conducted on real data confirmed the effectiveness of the proposed approach, demonstrating a reduction in forecast error due to preliminary clustering of time series. The developed solution can be useful in economics, medicine, industry, energy and other areas where high forecasting accuracy is required. The article will be useful for developers working with machine learning, time series analysis, as well as researchers studying modern methods of data mining and their practical application.
The article presents a software package for time series forecasting based on clustering and deep learning methods. The package includes a control controller, modules for data preprocessing, time series parameterization, clustering using the DBSCAN algorithm, and forecasting using LSTM recurrent neural networks. An important feature is the parameterization of time series before clustering, which allows you to highlight key statistical and spectral characteristics of the data, such as mean, variance, ARIMA coefficients, trend, number of peaks and troughs. The use of the DBSCAN clustering algorithm provides automatic formation of clusters without the need to specify their number in advance, which makes the algorithm more flexible and resistant to noise. The use of separate LSTM models for each cluster allows you to take into account specific data patterns within a group, increasing the accuracy of predictions. The implementation is made in Python and ASP.NET Core, which simplifies integration into various information systems and ensures cross-platform. The article discusses in detail the architecture of the software package, its functionality, data processing algorithms and forecasting methodology. Experiments conducted on real data confirmed the effectiveness of the proposed approach, demonstrating a reduction in forecast error due to preliminary clustering of time series. The developed solution can be useful in economics, medicine, industry, energy and other areas where high forecasting accuracy is required. The article will be useful for developers working with machine learning, time series analysis, as well as researchers studying modern methods of data mining and their practical application.
Keywords:
TIME SERIES FORECASTING, SOFTWARE PACKAGE, CLUSTERING, DBSCAN, LSTM, DATA PREPROCESSING, TIME SERIES PARAMETERIZATION, MACHINE LEARNING, ASP.NET CORE, PYTHON
TIME SERIES FORECASTING, SOFTWARE PACKAGE, CLUSTERING, DBSCAN, LSTM, DATA PREPROCESSING, TIME SERIES PARAMETERIZATION, MACHINE LEARNING, ASP.NET CORE, PYTHON
