Dynamic selection of k nearest neighbors in instance-based learning

Hulett, Carl; Hall, Andy; Qu, Guangzhi

doi:10.1109/iri.2012.6302995

Cited by 6 publications

(3 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Then, the profiles were applied as inputs to the advanced data classification methods: Multilayer Percepton, a feedforward ANN model of interconnected nodes that maps, in a non-linear way, the connections between the input and the output data [11], a more practical Support Vector Machine (SVM) algorithm that simplifies the calculations performed when the number of variables is higher [12], and Instance Based for K-Nearest neighbour (k-NN), that assigns each instance to a given group according to the similarity with the majority of its neighbour [13]. 10.21611/qirt.2018.013…”

Section: Methodsmentioning

confidence: 99%

Classifying Skin Neoplasms with Infrared Thermal Images

Magalhães¹,

Vardasca²,

Mendes³

2018

Proceedings of the 2018 International Conference on Quantitative InfraRed Thermography

View full text Add to dashboard Cite

The cases of benign and malignant skin neoplasms have been rising over the last decades. They are an esthetical and health threat and require prompt attention, the current diagnostic methods are the biopsy (expensive and invasive) and the dermatoscopy (experience dependent), so new objective methods are needed. The growth of neoplasms is influenced by blood flow, influencing, thereby, skin temperature. Infrared thermal images were taken from 85 patients at the cancer hospital, being analysed and classified with different intelligent data methods. The method that outperformed the other was the k-Nearest Neighbour with 60% accuracy.

show abstract

Section: Methodsmentioning

confidence: 99%

Classifying Skin Neoplasms with Infrared Thermal Images

Magalhães¹,

Vardasca²,

Mendes³

2018

Proceedings of the 2018 International Conference on Quantitative InfraRed Thermography

View full text Add to dashboard Cite

show abstract

“…This algorithm compares the similarity between data points and will put the input data in the same class as the data point that is closest to Hulett et al (2012); Piryonesi and El-Diraby (2020). The distance can be calculated by using a distant equation, e.g., the Euclidean distance (Hastie et al, 2009) as shown in formula 1, when x, y is the data coordinate and a, b is the target coordinate that we want to find the distance between them.…”

Section: K-nearest Neighbor Algorithmmentioning

confidence: 99%

Development of Customer Predictive Model for Investment Using Ensemble Learning Technique

Kaewkiriya¹,

Wisaeng²

2023

Journal of Computer Science

View full text Add to dashboard Cite

Many new investors that want to purchase a fund might be disappointed with the return-on-investment value. This problem occurs because they do not know important factors that could affect the market value. Plus, some assets might not be suitable for the investor's investment style. To reduce the mentioned problems, many asset management companies research the systems that could find a suitable fund for the investor. These systems consider various aspects, such as the investor's background, financial status, and the investor's behavior. These systems typically employ a specific machine learning method to learn and predict which fund the model should recommend to customers. However, we have an assumption that the performance of the prediction could be leveraged if we applied more methods to a forecasting model. Therefore, this study aims to develop a customer predictive model for investment using voting ensemble learning techniques. The model is used for recommending suitable funds and suitable risks for investing based on the investor's profile and comparing performance between 5 algorithms and 2 preprocessing approaches. Preprocessing approaches are clustering by date range which has an average accuracy of 62.24% and k-means clustering which has an average accuracy of 69.21%. The prediction model of suitable fund risk and prediction model of fund category has an accuracy of 92.38%. We found that the neural network has the highest accuracy of 93.43%.

show abstract

“…The label of an unlabeled instance is decided by the label belonging to most of the K neighbors which have nearest distance between them and the instance. 4 Traditional K-Means algorithm has many shortcomings, for instance, initial representative points are selected randomly and updating them is sensitive to noisy. 2 Clustering result is greatly affected by the cluster center selected randomly and is often trapped in local optimal solution rather than global optimal solution.…”

Section: Introductionmentioning

confidence: 99%

A Novel K-Means Clustering Algorithm and Its Application

Peng¹,

Li²,

Li³

2015

j comput theor nanosci

View full text Add to dashboard Cite

K -Means algorithm has been researched adequately in recent years. Clustering result of traditional K -Means algorithm is affected by the choice of initial point and noise. In addition to, traditional K -Means algorithm only favors clusters with spherical shapes and similar sizes. A novel K -Means algorithm combining K -Means algorithm and KNN algorithm called KK -Means is proposed to solve these weaknesses in this paper. Experimental result shows that KK -Means algorithm has better performance more than traditional K -Means algorithm.

show abstract

Dynamic selection of k nearest neighbors in instance-based learning

Cited by 6 publications

References 24 publications

Classifying Skin Neoplasms with Infrared Thermal Images

Classifying Skin Neoplasms with Infrared Thermal Images

Development of Customer Predictive Model for Investment Using Ensemble Learning Technique

A Novel K-Means Clustering Algorithm and Its Application

Contact Info

Product

Resources

About