Abstract-This paper presents a hybrid approach with two phases for improving the performance of training artificial neural networks (ANNs) by selection of the most important instances for training, and then reduction the dimensionality of features. The ANNs which are applied in this paper for validation, are included Multi-Layer Perceptron (MLP) and Neuro-Fuzzy Network (NFN). In the first phase, the Modified Fast Condensed Nearest Neighbor (MFCNN) algorithm is used to construct the subset with instances very close to the decision boundary. It leads to achieve the instances more useful for training the network. And in the second phase, an Ant-based approach to the supervised reduction of feature dimensionality is introduced, aims to reduce the complexity, and improve the accuracy of learning the ANN. The main purpose of this method is to enhance the classification performance by improving the quality of the training set. Experimental results illustrated the efficiency of the proposed approach.Index Terms-Artificial neural networks, instance selection, feature selection, ant colony optimization, EDA, MFCNN. I. INTRODUCTIONThe learning procedure has different steps including, constructing a Training Set (TS), training the network, testing the behavior of the network on the new instances. The first step consists of editing and condensing the data. The most important objective of any abstraction technique is to obtain a TS with the best performance in classification process. To build an optimal TS, we have to select the instances and features in an appropriate way to guarantee attaining to the best result.The whole data is located in a matrix which its rows show the instances and the columns are the features. The instance selection schemes generally try to reduce the number of rows in entire dataset with no loss in the classification performance [1]. Non-parametric K-Nearest Neighbor (KNN) rule [2], predicts the label of a new instance by computing a similarity scale between the selected one and all other instances in TS. Condensed Nearest Neighbor rule (CNN) [3], was introduced for the nearest neighbor classification method and tried to build a condensed subset from the whole set of examples. The basic idea of this algorithm is that if an example is not correctly classified, this may occur because it is a boundary instance, and it should be considered in the selected condense set.Fast set S that is initialized with centroid of each class in TS. Then, until the TS is not completely classified with condensed set S, the algorithm selects for insertion a representative of the misclassified points of each Voronoi (Voronoi of point P∈S is the set of all instances in TS that are closer to P than to any other instance in S) cell induced by the current subset. Edited Nearest Neighbor (ENN) removes noisy instances, as well as close border points, leaving smoother decision boundaries [5]. It edits out those prototypes that misclassified using KNN rule. Allknn [6] is based on the ENN method. This algorithm, for k=1 to m, marks...
This paper presents new method for training intelligent networks such as Multi-Layer Perceptron (MLP) and Neuro-Fuzzy Networks (NFN) with prototypes selected via Fast Condensed Nearest Neighbor (FCNN) rule. By applying FCNN, condensed subsets with instances close to the decision boundary are obtained. We call these points High-Priority Prototypes (HPPs) and the network is trained by them. The main objective of this approach is to improve the performance of the classification by boosting the quality of the training-set. The experimental results on several standard classification databases illustrated the power of the proposed method. In comparison to previous approaches which select prototypes randomly, training with HPPs performs better in terms of classification accuracy.
Cancer is one of the most dangerous diseases around the world and the most common cancer among women is breast cancer. Although not all the cancer types are curable upon diagnosis, breast cancer can be cured if it is diagnosed early. The most reliable way of diagnosing breast cancer is mammographic screening which can diagnose the disease 1.5 to 4 years before it is clinically diagnosed. Double Reading is the important diagnostic process in which two experts/radiologists should read the same mammogram image to make an accurate diagnosis. But this process is not a cost-effective approach for early detection of breast cancer. Computer-Aided Diagnosis (CAD) can act as the second expert and therefore one expert would be enough for breast cancer diagnosis. In this study, we use the data extracted from low-resolution as well as high resolution mammography images. The attributes extracted from mammographic images are imported into Support Vector Machine (SVM) to classify the patients. An important point about the attributes is that sometimes there may be some irrelevant or even noisy attributes that have negative effect on the classification accuracy. Therefore, the main objective of this study is to apply local and global search paradigms in order to find the best subset of attributes to construct the most accurate CAD system that can effectively distinguish between benign and healthy patients. Artificial Bee Colony (ABC) is a population-based swarm intelligence algorithm with good global exploration ability, and Simulated Annealing (SA) is a robust local-search algorithm. Thus, we utilize a hybrid global and local search algorithm (named ABCSA) to simultaneously benefit from the advantages of both ABC and SA. In this approach, ABC is firstly performed for the global exploration in the search space. Then, SA is utilized to search locally in the vicinity of the best solution found via ABC, in order to improve the quality of the final solution. Obtained simulation results over four different mammographic datasets show that the proposed algorithm outperforms the existing metaheuristic feature selection approaches in terms of minimizing the number of features, while maximizing the detection accuracy.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.