Search

Labeled data is the main ingredient for classification tasks. Labeled data is not always available and free. Semi-supervised learning solves the problem of labeling the unlabeled instances through heuristics. Self-training is one of the most widely-used comprehensible approaches for labeling data. Traditional self-training approaches tend to show low classification accuracy when the majority of the data is unlabeled. A novel approach named Self-Training using Associative Classification using Ant Colony Optimization (ST-AC-ACO) has been proposed in this article to label and classify the unlabeled data instances to improve self-training classification accuracy by exploiting the association among attribute values (terms) and between a set of terms and class labels of the labeled instances. Ant Colony Optimization (ACO) has been employed to construct associative classification rules based on labeled and pseudo-labeled instances. Experiments demonstrate the superiority of the proposed associative self-training approach to its competing traditional self-training approaches.

Discretization is the process of converting numerical values into categorical values. Contemporary literature study reveals that there are many techniques available for numerical data discretization. The performance of classification method is dependent on the exploitation of the data discretizing method. In this article, we investigate the effect of discretization methods on the performance of associative classifiers. Most of the classification approaches work on the discretized databases. There are various approaches exploited for the discretization of the database to compare the performance of the classifiers. The selection of the discretization method greatly influences the classification performance of the classification method. We compare the performance of associative classifiers namely CBA and CBA2 on the selective discretizing methods i.e. 1R Discretizer (1R-D), Ameva Discretizer (Ameva-D), Bayesian Discretizer (Bayesian-D), Discretization algorithm based on Class-Attribute Contingency Coefficient (CACC-D), Class-Attribute Dependent Discretizer (CADD-D), Distribution-Index-Based Discretizer (DIBD-D), Cluster Analysis (ClusterAnalysis-D), Chi-Merge Discretizer (ChiMerge-D) and Chi2 Discretizer (Chi2-D) in terms of accuracy. The main object of this study is to investigate the impact of discretizing method on the performance of the Associative Classifier by keeping constant other experimental parameters. Our experimental results show that the performance of the Associative Classifier significantly varies with the change of data discretization method. So the accuracy rate of the classifier is highly dependent on the selection of the discretization method. For this comparative performance study, we use the implementation of these methods in KEEL data mining tool on public datasets.

In this paper, a novel approach for discovering emerging patterns has been proposed. Majority of the existing algorithms for the discovery of emerging patterns are tree-based which involve growth and shrinking of trees for this purpose. These algorithms follow greedy search approach for discovery of emerging patterns. The proposed approach utilizes the diversity of ant colony optimization and avoids complexity and greedy search of tree-based algorithms for discovery of emerging patterns. The experiments show that the proposed approach provides higher accuracy than existing state of the art classifiers as well as emerging pattern-based classifiers.

This article represents the focused performance analysis of the statistical Pattern Recognition methods. There is a variety of emerging pattern recognition problems for different data representations like multimedia, spatial, temporal and textual data. Statistical Pattern Recognition approaches are extensively applied for pattern recognition and classification purposes. There are various statistical pattern-recognition methods proposed in the contemporary literature. For the selection of more appropriate statistical pattern recognition, method demands comprehensive performance evaluation of contemporary methods. However, this empirical study is focused on the performance evaluation of statistical methods for pattern discovery in terms of accuracy and error rate. The studied methods include Naive-Bayes (NB-C), Linear Discriminant Analysis (LDA-C), Kernel Classifier (Kernel-C), Least Mean Square Linear Classifier (LinearLMS-C), Least Mean Square Quadratic classifier(PolQuadraticLMS-C), Multinomial logistic regression model with a ridge estimator(Logistic-C) and Particle Swarm Optimization - Linear Discriminant Analysis (PSOLDA-C). The implementation of mentioned methods in KEEL, a data mining tool, is applied on public datasets for the comparative analysis. Experimental results reveal that the performance of PSOLDA-C is promising than other methods in terms of accuracy.

Clustering plays a major role in data mining. It helps in identifying patterns and distribution of data. In this paper, we propose ant colony optimization (ACO) based clustering algorithm for clustering the social network data. The proposed technique takes advantage of Ants cemetery and allows each ant to play a role. In every iteration, the ants produce a cluster for the data, and the pheromone values are updated after every iteration of all the ants. The experiment results show that proposed technique discovered the clusters which reveal the truthfulness of the network. We have also compared the proposed technique with another state of the art clustering algorithm, and experimental result demonstrated that proposed technique find out the better clusters.

Ant-Miner, a rule-based classification algorithm, has been successfully applied for classification tasks but it has some limitations such as getting stuck in local optima, high selective pressure, fixed exploration and exploitation rate, and premature convergence. In this paper, we have proposed a novel Ant-Miner based technique based on new Pheromone update method, Rule Rejection threshold, Adaptive gamma, and altered Tournament selection (PRRAT_AM) that caters to these limitations. The proposed algorithm introduced an adaptive gamma parameter to avoid fixed exploration and exploitation rate. To decrease the selective pressure, pheromone is updated by weighted average of rule length, rule quality and heuristic of the path. Ants are selected using improved tournament selection strategy to update the pheromone. Rules that covered less than one percent of the training examples are rejected to generate generic rules. These improvements aid PRRAT_AM in avoiding premature convergence and high selective pressure. We have tested the proposed approach on eight publicly available data-sets on standard benchmark performance measures that include accuracy and F1-score. The proposed approach has been compared with state of the art versions of Ant-Miner and with various data mining algorithms. The experimental results showed that the proposed approach achieved better results when compared with other techniques in terms of standard performance measures and convergence speed.

Clustering is a popular machine learning technique used to segregate the number of objects into different classes. Locating multiple peaks of a multimodal problem is a challenging task. In this regard, clustering is considered to be the best technique that divides the population into a number of clusters. In literature, a number of clustering techniques have been suggested with the particle swarm optimization (PSO) algorithm. Among different clustering techniques, partition-based clustering is widely used with the PSO to solve multimodal optimization problems. This paper provides a brief survey of clustering-based PSO techniques for solving multimodal optimization problems. Open problems have been discussed at the end of the paper as well.

Data is an important asset for any organization to successfully run its business. When we collect data, it contains data with low qualities such as noise, incomplete, missing values etc. If the quality of data is low then mining results of any data mining algorithm will also below. In this paper, we propose a technique to deal with missing values. Genetic algorithm (GA) is used for the estimation of missing values in datasets. GA is introduced to generate optimal sets of missing values and information gain (IG) is used as the fitness function to measure the performance of an individual solution. Our goal is to impute missing values in a dataset for better classification results. This technique works even better when there is a higher rate of missing values or incomplete information along with a greater number of distinct values in attributes/features having missing values. We compare our proposed technique with single imputation techniques and multiple imputations (MI) statistically based approaches on various benchmark classification techniques on different performance measures. We show that our proposed methods outperform when compare with another state of the art missing data imputation techniques.

In this research study, we analyze the performance of bio inspired classification approaches by selecting Ant-Miners (Ant-Miner, cAnt_Miner, cAnt_Miner2 and cAnt_MinerPB) for the discovery of classification rules in terms of accuracy, terms per rule, number of rules, running time and model size discovered by the corresponding rule mining algorithm. Classification rule discovery is still a challenging and emerging research problem in the field of data mining and knowledge discovery. Rule based classification has become cutting edge research area due to its importance and popular application areas in the banking, market basket analysis, credit card fraud detection, costumer behaviour, stock market prediction and protein sequence analysis. There are various approaches proposed for the discovery of classification rules like Artificial Neural Networks, Genetic Algorithm, Evolutionary Programming, SVM and Swarm Intelligence. This research study is focused on classification rule discovery by Ant Colony Optimization. For the performance analysis, Myra Tool is used for experiments on the 18 public datasets (available on the UCI repository). Data sets are selected with varying number of instances, number of attributes and number of classes. This research paper also provides focused survey of Ant-Miners for the discovery of classification rules.

There exist numerous state of the art classification algorithms that are designed to handle the data with nominal or binary class labels. Unfortunately, less attention is given to the genre of classification problems where the classes are organized as a structured hierarchy; such as protein function prediction (target area in this work), test scores, gene ontology, web page categorization, text categorization etc. The structured hierarchy is usually represented as a tree or a directed acyclic graph (DAG) where there exist IS-A relationship among the class labels. Class labels at upper level of the hierarchy are more abstract and easy to predict whereas class labels at deeper level are most specific and challenging for correct prediction. It is helpful to consider this class hierarchy for designing a hypothesis that can handle the tradeoff between prediction accuracy and prediction specificity. In this paper, a novel ant colony optimization (ACO) based single path hierarchical classification algorithm is proposed that incorporates the given class hierarchy during its learning phase. The algorithm produces IF-THEN ordered rule list and thus offer comprehensible classification model. Detailed discussion on the architecture and design of the proposed technique is provided which is followed by the empirical evaluation on six ion-channels data sets (related to protein function prediction) and two publicly available data sets. The performance of the algorithm is encouraging as compared to the existing methods based on the statistically significant Student's t-test (keeping in view, prediction accuracy and specificity) and thus confirm the promising ability of the proposed technique for hierarchical classification task.

The primary objective of this research is to propose and investigate a novel ant colony optimization-based classification rule discovery algorithm and its variants. The main feature of this algorithm is a new heuristic function based on the correlation between attributes of a dataset. Several aspects and parameters of the proposed algorithm are investigated by experimentation on a number of benchmark datasets. We study the performance of our proposed approach and compare it with several state-of-the art commonly used classification algorithms. Experimental results indicate that the proposed approach builds more accurate models than the compared algorithms. The high accuracy supplemented by the comprehensibility of the discovered rule sets is the main advantage of this method.

This paper presents a novel high speed clustering scheme for high-dimensional data stream. Data stream clustering has gained importance in different applications, for example, network monitoring, intrusion detection, and real-time sensing. High dimensional stream data is inherently more complex when used for clustering because the evolving nature of the stream data and high dimensionality make it non-trivial. In order to tackle this problem, projected subspace within the high dimensions and limited window sized data per unit of time are used for clustering purpose. We propose a High Speed and Dimensions data stream clustering scheme (HSDStream) which employs exponential mov-ing averages to reduce the size of the memory and speed up the processing of projected subspace data stream. It works in three steps: i) initialization, ii) real-time maintenance of core and outlier micro-clusters, and iii) on-demand offline generation of the final clusters. The proposed algorithm is tested against high dimensional density-based projected clustering (HDDStream) for cluster purity, memory usage, and the cluster sensitivity. Experi-mental results are obtained for corrected KDD intrusion detection dataset. These results show that HSDStream outperforms the HDDStream in all performance metrics, especially, the memory usage and the processing speed.

Network security risks grow with increase in the network size. In recent past, the attacks on computer networks have increased tremendously and require efficient network intrusion detection mechanisms. Data mining and machine-learning techniques have been used for network intrusion detection during the past few years and have gained much popularity. In this paper, we propose an intrusion detection mechanism based on binary particle swarm optimization PSO and random forests RF algorithms called PSO-RF and investigate the performance of various dimension reduction techniques along with a set of different classifiers including the proposed approach. Binary PSO is used to find more appropriate set of attributes for classifying network intrusions, and RF is used as a classifier. In the preprocessing step, we reduce the dimensions of the dataset by using different state-of-the-art dimension reduction techniques, and then this reduced dataset is presented to the proposed PSO-RF approach that further optimizes the dimensions of the data and finds an optimal set of features. PSO is an optimization method that has a strong global search capability and is used here for dimension optimization. We perform extensive experimentation to prove the worth of the proposed approach by using different performance metrics. The standard benchmark, that is, KDD99Cup dataset, is used that contains the information about various kinds of network intrusions. The experimental results indicate that the proposed approach performs better than the other approaches for the detection of all kinds of attacks present in the dataset. Copyright © 2012 John Wiley & Sons, Ltd.

In recent years, a few sequential covering algorithms for classification rule discovery based on the ant colony optimization meta-heuristic (ACO) have been proposed. This paper proposes a new ACO-based classification algorithm called AntMiner-C. Its main feature is a heuristic function based on the correlation among the attributes. Other highlights include the manner in which class labels are assigned to the rules prior to their discovery, a strategy for dynamically stopping the addition of terms in a rule’s antecedent part, and a strategy for pruning redundant rules from the rule set. We study the performance of our proposed approach for twelve commonly used data sets and compare it with the original AntMiner algorithm, decision tree builder C4.5, Ripper, logistic regression technique, and a SVM. Experimental results show that the accuracy rate obtained by AntMiner-C is better than that of the compared algorithms. However, the average number of rules and average terms per rule are higher.

Ant Miner algorithms are used for the classification task of data mining. This paper proposes a new version of Ant Miner algorithm called CAnt Miner that considers compatibility between the attributes, when constructing a rule. In addition to a new heuristic function based on compatibility among attributes, we also introduce a strategy to dynamically stop adding terms in the rule antecedent part instead of using a user specified threshold that specifies when to stop adding terms. We compare the performance of proposed approach with other Ant Miner versions and C4.5 in nine public domain data set. The experiment results show that the proposed approach achieves higher accuracy rate then other approaches.

Association rules mining and classification rules discovery are two important data mining techniques used to expose the relations among large sets of data items. The technique aims to find out the rules that satisfy the predefined minimum support and the confidence. Association rules mining has successfully been implemented in biomedical research and has demonstrated encouraging results in analysing the gene expression data in order to discover the relevant biological association among different genes, gene expression, and various protein properties like protein functionality and sequence similarity. In this paper, we applied the association rule mining technique - the ACO-AC to the problem of classifying proteins into its correct fold of the SCOP dataset. The technique combines the association rules mining and supervised classification mechanism using ant colony optimisation. Experimental results reveal the classifier performance in protein classification problem as excellent by identifying most accurate and compact rules.

Feature subset selection is one of the key problems in the area of pattern recognition and machine learning. Feature subset selection refers to the problem of selecting only those features that are useful in predicting a target concept i.e. class. Data acquired through different sources are not particularly screened for any specific task e.g. classification, clustering, anomaly detection, etc. When the data are fed to a learning algorithm, its results deteriorate. The proposed method is a pure filter based feature subset selection technique which incurs less computational cost and highly efficient in terms of classification accuracy. Moreover, along with high accuracy the proposed method requires less number of features in most of the cases. In the proposed method the issue of feature ranking and threshold value selection is addressed. The proposed method adaptively selects number of features as per the worth of an individual feature in the dataset. An extensive experimentation is performed, comprised of a number of benchmark datasets over three well known classification algorithms. Empirical results endorse efficiency and effectiveness of the proposed method.

Nowadays, size of databases is increasing drastically which requires huge memory and high computational power to overcome memory and computational limitations efficiently. To increase performance and overcome memory limitation we need distributed approach. In this paper, a three step distributed approach is proposed which divides the large data sets into data chunks initially, processes it on defined N processors on different machines, generates the final merged decision rule file and resolves the conflicts that may arise later on. Mostly, classification algorithms generates only specific or generic decision rules, in contrast to traditional algorithms proposed solution has capability to generate both specific and generic rules. This approach shows promising results in terms of accuracy and efficiency and well suited for distributed environment.

During the past few years, huge amount of network attacks have increased the requirement of efficient network intrusion detection techniques. Different classification techniques for identifying various attacks have been proposed in the literature. In this paper we propose and implement a hybrid classifier based on binary particle swarm optimization (BPSO) and random forests (RF) algorithm for the classification of PROBE attacks in a network. PSO is an optimization method which has a strong global search capability and is used for fine-tuning of the features whereas RF, a highly accurate classifier, is used here for classification. We demonstrate the performance of our technique using KDD99Cup dataset. We also compare the performance of our proposed classifier with eight other well-known classifiers and the results show that the performance achieved by the proposed classifier is much better than the other approaches.

In this work, we propose a Comprehensive Learning Particle Swarm Optimization (CLPSO) based weighted clustering algorithm for mobile ad hoc networks. It finds the optimal number of clusters to efficiently manage the resources of the network. The proposed CLPSO based clustering algorithm takes into consideration the ideal degree, transmission power, mobility, and battery power of the mobile nodes. A weight is assigned to each of these parameters of the network. Each particle contains information about the cluster-heads and the members of each cluster. The simulation results are compared with two other well-known clustering algorithms. Results show that the proposed technique works better than the other techniques especially in dense networks.