Algorithm for the Development of a Training Set that Best Describes the Objects of Recognition

Kamilov, Mirzoyan; Hudayberdiev, Mirzaakbar; Alisher, Khamroev

doi:10.1016/j.procs.2019.02.024

Cited by 6 publications

(1 citation statement)

References 7 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…У статті [4] представлено новий непараметричний лінійний метод вилучення ознак для класифікації гіперспектральних зображень, який використовує ідеї вікон Парзена для визначення локального середнього значення сусідніх зразків і нові вагові функції для формування матриць розкиду міжкласової та внутрішньокласової варіації. У статті [3] запропоновано алгоритм роз-роблення навчального набору, який найкраще описує об'єкти розпізнавання.…”

Section: огляд літературиunclassified

Method for Forming Training Samples for Data Arrays Based on Machine Learning

Lipianina-Honcharenko

2024

NRPCOMP

View full text Add to dashboard Cite

The study introduces an innovative methodology for crafting training samples through the integration of machine learning techniques. This method encompasses a fusion of RFM (Recency, Frequency, Monetary) analysis and cluster analysis, offering a comprehensive approach to sample formation. The application of this approach is demonstrated on a dataset derived from concluded tender agreements by participants in Ukraine, sourced from the ProZorro Sales platform. The compiled dataset encompasses an impressive volume, encompassing a total of 92,638 auctions, which further breaks down into 29,164 distinct auctions and an assemblage of 39,747 unique organizers.The utilization of RFM analysis within this framework yields the categorization of the dataset into distinct groups, each characterized by its own distinct attributes. These groupings include designations such as “The Best Organizers of Tenders,” “Loyal Organizers of Tenders,” “Large Consumers,” “Tenders Held Infrequently but with Substantial Sums,” and “Weak Tender Organizers.” Following the RFM analysis, the K-means clustering methodology is implemented, resulting in the division of the data into five clusters, each contributing to a nuanced differentiation of diverse organizer profiles.Intriguingly, a comparative analysis involving RTF (Relative Total Frequency) scores and the K-means groupings reveals congruence between clusters representing organizers who actively orchestrate numerous tenders with significant monetary value, as well as clusters characterized by minimal tender activity with less substantial monetary implications. To validate the efficacy of the proposed method, rigorous testing is conducted employing Logistic Regression and Naive Bayes algorithms. Encouragingly, the results consistently showcase impressive accuracy for both methods, highlighting their robustness.An outlook towards future research endeavors suggests a promising avenue of developing an automated system for the selection of tender organizers, underpinned by machine learning principles. Such a system would undoubtedly revolutionize the optimization of participation strategies within the domain of tender processes, fostering efficiency and accuracy in decision-making.

show abstract

Section: огляд літературиunclassified