| Autor: |
Tajmouati, Samya, El Wahbi, Bouazza, Dakkon, Mohamed |
| Zdroj: |
OPSEARCH; Sep2023, Vol. 60 Issue 3, p1063-1081, 19p |
| Abstrakt: |
In this paper, we study the convergence of the algorithms employed by Classical Parameters Tuning in Nearest Neighbors (CPTO-WNN) and Fast Parameters Tuning in Nearest Neighbors (FPTO-WNN). CPTO-WNN and FPTO-WNN are two methodologies that perform the selection of weighted nearest neighbors parameters in time series setting. To do so, each methodology employs an algorithm named WNNoptimization. The WNNoptimization algorithm employs a method inspired by time series cross validation to pick automatically the weighted nearest neighbors parameters. Practically, it computes the global accuracy measure over the test sets, for different values of weighted nearest neighbors parameters and returns the ones for which the accuracy measure is minimal. Compared to CPTO-WNN, FPTO-WNN presents the advantage of reducing time complexity. However, when we are faced with many iterations, both methods raise concerns about computational time. One solution is to introduce a stop condition if the algorithms are convergent. Real data examples on retail and food services sales in the USA and milk production in the UK are analyzed to demonstrate the convergence of the algorithms. As a result, we propose a stop condition allowing the reduction of time complexity while maintaining a good precision. We compare the forecasting performance and time complexity of CPTO-WNN and FPTO-WNN after implementing the proposed stop condition. Experiments show that the introduction of the proposed stop condition provides good accuracy along with lower computational time. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink