Використання мiр подiбностi в методах класифiкацiї
| Jazyk: | angličtina |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Measure (data warehouse)
Computer science business.industry Process (computing) knn міра подібності Function (mathematics) Object (computer science) computer.software_genre алгоритм k найближчих сусідів Class (biology) ComputingMethodologies_PATTERNRECOGNITION Software Data extraction Similarity (network science) QA1-939 класифікація Data mining business computer Mathematics контрольоване машинне навчання |
| Zdroj: | Науковий вісник Ужгородського університету. Серія: Математика і інформатика, Vol 38, Iss 1, Pp 143-148 (2021) |
| ISSN: | 2616-7700 |
| Popis: | This study is a development of the application of different types of similarity measures in data mining problems. Data mining is the process of extracting implicit information from a database, which characterizes hidden connections and structures. This type of analysis is projected to be extremely popular over the next decade. One of the methods of data extraction is classification. The paper provides an overview of modern areas of supervised classification. The most popular method of classifying objects with numerical attributes is the method of K-nearest neighbors (KNN). It has been found that the predictive value of a class label can be improved by using the weighted influence of each neighbor on the result. Thus, it is advisable to modify the KNN method. In this case, it is proposed to introduce a function that characterizes the similarity of the unlabeled object with its nearest neighbors in the form of a measure of similarity. Based on it, indicators of weighted counting of votes of neighbors for a certain class mark are introduced. Software has been developed that implements the described approach. Practical experiments have shown its effectiveness in solving certain classes of applied problems. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|