Iraqi Journal of Intelligent Computing and Informatics

Particle Swamp Optimization (PSO) is an effective method for solving a wide range of problems. However, the most existing PSO algorithms easily trap into local optima when solving complex multimodal function optimization problems. In this paper, we explain the importance of PSO algorithm’s general purposes to optimize strategy which has various parameters that decide its conduct and viability in advancing agiven issue. This study gives a rundown of the best selections of parameters for different advancement situations which should enable the specialist to accomplish better outcomes with less exertion. In this paper, we define an important in the PSO algorithm, the parameters and how to apply this algorithm to different type of datasets online and offline. Repeat Consumption Matrices Dataset has been tested with PSO algorithm to define the number of clusters using selection of parameters. The results are thenhaving been compared with the Genetic Algorithm (GA) where the algorithms have been tested based on the number of cluster and DB index for each of the data. Based on the results, it is shown that the PSO algorithm gives good and efficient results comparedto GA algorithm. The output of the research is a PSO-based clustering algorithm that can be used in data mining by providing accurate and robust results in clustering, which can be realized in web search engines and automatic document organization.

The advanced of Information Technology has resulting in the generation of numerous datasets with different dimensions. However, dealing with multi-dimensional datasets which typically contain large number of attributes, p has cause problems to classification process. Classifying different dimensional numerical data is a difficult problem as dealing with various feature spaces, could cause the performance of supervised learning method to suffer from the curse of dimensionality. This condition eventually degrades both classification accuracy and efficiency. In a nutshell, not all attributes in the dataset can be used in the classification process since some features may lead to low performance of classifier. Feature selection (FS) is a good mechanism that minimises the dimensions of high-dimensional datasets and solve classification problems. This paper proposed Bat Algorithm (BA) for FS that were trained using a Support Vector Machine (SVM) classifier. The proposed algorithm was tested on six public datasets with different sizes and compared with other benchmark algorithms, such as Particle Swarm Optimisation (PSO) and Genetic Algorithm (GA). The experimental results indicated that the BA has outperformed the other two algorithms. In addition, the comparison details showed that binary BA is more competitive in terms of accuracy and the number of features when assessed on datasets with different sizes.

Intrusion Detection Systems (IDSs) are efficient applications that monitor activities of specific network or system to detect any abnormal activity and then send alarms for a defined management station. However, the current IDSs generate a high number of false alarms; False Positives (FP) and False Negatives (FN), which decreases the accuracy of distinguishing attacks from normal activities. Therefore, this thesis introduces the implementation of enhanced IDS using two classifiers: PCA-SVM and PCA-KNN. The performance of the system with using these classifiers is compared using the NSL-KDD dataset to determine the optimal classifier in terms of detection rate and the number of generated false alarms. This is performed based on dividing the dataset into training and testing sets, where the Control Chart is then applied on the training set to improve the results, where it filters the data to remove the out-bound data and keep the data in the range from Mean-3sigma to Mean+3sigma.

Six evaluation metrics; FP, FN, True Positive (TP), True Negative (TN), Detection Rate (DR) and Classification Rate (CR) are computed for both classifiers for three sets of features; with and without applying a control chart. The obtained results demonstrate that the PCA-KNN based IDS with control chart offers the best detection rate with minimum number of generated false alarms for sets F2 and F3, while the PCA-SVM based IDS with control chart offers the best detection rate with minimum number of generated false alarms for F1. The average achieved detection rate for the PCA-KNN based IDS is 98.17% with control chart and 88.7738% without control chart. On the other hand, the average achieved detection rate for the PCA-SVM based IDS is 97.62% with control chart and 96.63587% without control chart. Based on these outcomes, the application of control chart enhances the detection rate and decreases the number of false alarms for both classifiers. In addition, the PCA-KNN is the best classifier to be applied on the IDS with minimum number of false alarms and highest security and detection rate. Our proposed IDSs are implemented and tested in MATLAB 2014.

Many real-world applications, like multimedia retrieval, confront the difficulty of determining the distance between any two items on Multi-modal data.According to most current Dimension Metrics Learning (DML) techniques, distance metrics may be learned using just one feature type or an aggregated feature space where many features are simply connected. Even though DML has been extensively researched. This study proposes a new framework for online learning, as well as a new classroom learning system for online Multi-modal dimension metric learning (OMDML), that is both efficient and scalable.This paper proposes a low-rank OMDML calculation to reduce the expensive cost of DML on high-dimensional component space, which reduces the computational cost while maintaining extremely competitive or much higher learning accuracy.With the purpose of determining whether or not multi-modular image recovery calculations can be successfully implemented, a large number of experiments are carried out.In most datasets tested, the suggested approach consistently outperforms alternative state-of-the-art algorithms, according to extensive experimental results.