An Enhanced Performance of K-Nearest Neighbor (K-NN) Classifier to Meet New Big Data Necessities

Abstract

The rapid increase in the growth of text information over the past two decades has led to the need for the use of text classification techniques, particularly in the area of information retrieval, data mining and data management. The precise results and simplicity of the K-Nearest Neighbor Classification Algorithm (K-NN) in knowledge mining is the reason that made it one of the most important classification algorithms used in many tasks such as pattern recognition, regression, and text classification. Through experiments and analysis of the results of the use of the traditional algorithm of the (K-NN), there are some deficiencies in their performance, especially when the data are large such as the algorithm was unable to process big data by rapid extraction with minimal storage space and generate useless samples computation and probability problems. In this paper, we have developed an enhanced algorithm and get the best results and perform better than that in the traditional algorithm. The significant improvement in our model performance is due to the improvement by removing unnecessary computational samples in the traditional algorithm. The performance is further improved by using the lost value computational method to define results as a prelude to avoid wasting time by correcting and filtering noise, examining the database, and eliminating unwanted records. Additionally, the inverse logarithmic function was used to solve the probability problems the algorithm encounters. The experimental results showed the efficiency of the modified algorithm in reducing the sample size and speeding up the search for the required data.

Implementation of K-Nearest Neighbor using the oversampling technique on mixed data for the classification of household welfare status

“…KNN classification requires a distance that is in line with the type of research data (Wu et al, 2008). Several papers on the KNN technique and its development included Alsammak et al (2020) conducting research to improve the performance of the K-Nearest Neighbor (KNN) classifier to satisfy emerging big data requirements. Kirtania et al (2020) Wilson & Martinez (1997a) explained that one of the distances for numerical and categorical data types is the Heterogeneous Euclidean-Overlap Metric (HEOM).…”

Untitled

“…KNN classification requires a distance that is in line with the type of research data (Wu et al, 2008). Several papers on the KNN technique and its development included Alsammak et al (2020) conducting research to improve the performance of the K-Nearest Neighbor (KNN) classifier to satisfy emerging big data requirements. Kirtania et al (2020) Wilson & Martinez (1997a) explained that one of the distances for numerical and categorical data types is the Heterogeneous Euclidean-Overlap Metric (HEOM).…”

Artificial intelligence support for 5G/6G-enabled Internet of Vehicles networks: An overview

“…Similarly, the adoption of NB techniques in 5G/6Genabled IoV networks have their limitations in terms of the NB techniques' specifications which range from the assumption that the existence of a given characteristic in a class is completely different to the existence of any other characteristic to the "Zero frequency" case. Therefore, NB techniques need to individually treat the features and cannot extract useful details from the correlations among its specifications [64]. However, NB techniques can perform correctly in application areas where samples have similar and correlated features.…”