KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 10 of 10

Some of the metrics are blocked by your
consent settings
A Self-Adaptive Weights for K-Means Classification Algorithm
(2025-09-01)
Chenghu, Cui
;
Thammano, Arit
This paper presents an improved K-means clustering algorithm that addresses the traditional algorithm’s sensitivity to outlier and susceptibility to local optima by introducing an adaptive weight adjustment mechanism. It employs an exponential decay function to dynamically reduce the feature weights of outlier data points, effectively suppressing outliers while preserving the structure of the normal data. The proposed method retains the computational efficiency of standard K-means. Key contributions include: (a) A novel distance-based weighting strategy that progressively reduces the influence of noisy points, mitigating the impact of outliers on clustering performance. (b) An innovative form of "local dimensionality reduction" for outlier points via weight decay, which interferes only with the feature space of noisy regions while preserving the global topological structure of clean data. Extensive experiments on three benchmark datasets Iris (4-dimensional, balanced classes), Wine (13-dimensional, correlated features), and Wisconsin Breast Cancer Diagnosis (30-dimensional, imbalanced data) demonstrate the effectiveness of the approach. Compared to standard K-means, the proposed algorithm achieves accuracy improvements of 7.47% on Iris, 13.89% on Wine, and 19% on WBCD. This adaptive strategy offers a practical and efficient solution for clustering in noisy, high-dimensional environments, without the added complexity of mixture models.
Some of the metrics are blocked by your
consent settings
Backpropagation Neural Network with Adaptive Learning Rate for Classification
(2023-01-01)
Jullapak, Rujira
;
Thammano, Arit
This research aims to improve the classification accuracy by modifying an original backpropagation neural network. In the proposed BPNN-ZMP, the learning rates were automatic tuned to improve the classification accuracy. Breast Cancer Coimbra dataset and Banknote Authentication dataset were used for testing the model performances. The results demonstrate that BPNN-ZMP improved over the original backpropagation neural network by 12.12 and 11.46% for Breast Cancer Coimbra dataset and Banknote Authentication dataset respectively. Although BPNN-ZMP could improve the model accuracy, the high accuracy in neural network backpropagation has been challenged in future work.
Some of the metrics are blocked by your
consent settings
Deterministic Initialization of k-means Clustering by Data Distribution Guide
(2022-01-01)
Sirikayon, Chaloemphon
;
Thammano, Arit
Clustering by the k-means is the most widely used method because of its ease of use. But the disadvantage of the k-means algorithm is that it relies on a random initialization. Therefore, the results obtained from each clustering are not stable depending on the starting point, affecting the results obtained in other applications. This paper, therefore, presents a method for determining the initialization of the k-means algorithm using the Data Distribution Guide (DDG). And use it as an aid in determining the starting point without random. Make the results of clustering always equal. And from the experimental results, We found that the accuracy obtained from clustering using the initialization from this method was good. Compared to the commonly used initialization designation.
Some of the metrics are blocked by your
consent settings
Adaptive Learning Rate For Neural Network Classification Model
(2022-01-01)
Jullapak, Rujira
;
Thammano, Arit
;
Surakratanasakul, Boonprasert
Imbalanced data cause prediction inaccuracy of the classification model. Two types of techniques have been devised to address this problem: pre-processing data before training a classification model and adjusting the classification algorithm. This study, which introduced the adaptive learning rate into a backpropagation neural network algorithm, is of the latter type. The learning rate was adjusted in each iterative learning cycle: the learning rate is increased for the data class with fewer samples and decreased for the data class with more samples. K-fold cross-validation was used to test the effectiveness of the prediction model on 10 datasets. The results showed that the proposed ZMP algorithm outperformed the original backpropagation neural network on 6 datasets; the improvement ranged from 2.24% to 20.22%. Moreover, on the other 4 datasets, even though the proposed technique provided less accurate predictions, the differences were very slight.
Some of the metrics are blocked by your
consent settings
Adaptive Learning Rate for Dealing with Imbalanced Data in Classification Problems
(2021-03-03)
Jantanasukon, Ratanon
;
Thammano, Arit
This research modified a backpropagation learning algorithm in order to increase its ability to deal with imbalanced data problems. We used the backpropagation algorithm and a concept of multiple adaptive learning rates to train the feedforward neural network. Using multiple adaptive learning rates allowed us to achieve a classification model that had fewer problems when dealing with an imbalanced dataset the experimental results showed that the proposed method performed significantly better than the conventional backpropagation neural network in all tests.
Some of the metrics are blocked by your
consent settings
An improved invasive weed algorithm with RBFNN for optimizing classification problems
(2019-02-01)
Tansui, Daranat
;
Thammano, Arit
Classification or decision-making task is very common in every field of endeavor. Many algorithms have been developed to help to make this kind of task successful. Invasive Weed Optimization (IWO) has been commonly used to help perform this task. However, its classification accuracy still leaves something to be desired. This study attempted to develop an improved Invasive Weed Optimization (IIWO) that would perform this task better. Our objectives were as follows: 1) to use IWO to optimize the parameters of Radial Basis Function Neural Network classifier and 2) to improve the dispersal of offspring solutions in the search space by using 3 spatial distributions: Normal, Cauchy, and Levy distributions, instead of only one as in IWO. We evaluated the performance of IIWO against two conventional classification algorithms: Genetic Algorithm (GA) and Gradient Descent (GD) algorithm on 5 benchmark datasets and found that IIWO made more accurate predictions than these two algorithms on 3 out of the 5 datasets and nearly as accurate as them on the other 2 datasets. The reasons that IIWO performed as well as this was that the inner working of RBFNN allowed the algorithm to estimate the objective value more accurately and the use of 3 spatial distributions to disperse offspring solutions instead of one in IWO helped make the offspring solutions spread to cover the whole search space better.
Some of the metrics are blocked by your
consent settings
A new computational intelligence technique based on human group formation
(2010-03-01)
Thammano, Arit
;
Moolwong, Jittraporn
This paper proposes a novel computational intelligence technique, based on the sociological concept of human group formation, with the aim to acquire a better solution to classification problems. The key concept of the human group formation is about the behavior of in-group members that try to unite with their own group as much as possible, and at the same time maintain social distance from the out-group members. This study compares the performance of the proposed model with that of fuzzy ARTMAP, radial basis function network, and learning vector quantization. Experimental results demonstrate the potential of the proposed approach in offering an efficient and effective solution to the problem. © 2009 Elsevier Ltd. All rights reserved.
Some of the metrics are blocked by your
consent settings
Feedforward neural network with multi-valued connection weights
(2009-09-11)
Thammano, Arit
;
Ruxpakawong, Phongthep
This paper introduces a new concept of the connection weight to the multi-layer feedforward neural network. The architecture of the proposed approach is the same as that of the original multi-layer feedforward neural network. However, the weight of each connection is multi-valued, depending on the value of the input data involved. The backpropagation learning algorithm was also modified to suit the proposed concept. This proposed model has been benchmarked against the original feedforward neural network and the radial basis function network. The results on six benchmark problems are very encouraging. © 2009 Springer Berlin Heidelberg.
Some of the metrics are blocked by your
consent settings
Solving classification problems using supervised self-organizing map
(2007-12-01)
Thammano, Arit
;
Kiatwuthiamorn, Jiraporn
This paper proposes the new approach to deal with the classification problems by modifying the well-known Kohonen self-organizing map in order to make it able to solve classification problems. During training, the fuzzy membership function is used in place of the Euclidean distance to find the best matching cluster for the input pattern. In order to improve the efficiency of proposed model, the fuzzy entropy concept is employed to reduce the number of nodes in the cluster layer. The performance of the proposed model was compared with the fuzzy ARTMAP neural network. The results on five benchmark problems are very encouraging. ©2007 IEEE.
Some of the metrics are blocked by your
consent settings
Speech recognition of thai digits using modified cross-correlation neural network
(2004-12-01)
Thammano, Arit
;
Klomiam, Narodom
In this paper, Modified Cross-Correlation Neural Network (MCCNN), which is an extension of Cross-Correlation Neural Network (CCNN) [1], is proposed. Unlike the CCNN, which utilizes the normalized cross-correlation at zero lag as a choice function to determine the winning cluster node, MCCNN uses the maximum of the normalized cross-correlation instead. In this work, spoken Thai digits (0-9) are used as the experimental data. The performance of MCCNN, CCNN and other two well-known algorithms, Back-propagation and Fuzzy ARTMAP, are compared. The results show that MCCNN has the best performance with respect to the recognition rate.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results