Noorhaniza Wahid

A hybrid network intrusion detection system using simplified swarm optimization (SSO)

The network intrusion detection techniques are important to prevent our systems and networks from malicious behaviors. However, traditional network intrusion prevention such as firewalls, user authentication and data encryption have failed to completely protect networks and systems from the increasing and sophisticated attacks and malwares. In this paper, we propose a new hybrid intrusion detection system by using intelligent dynamic swarm based rough set (IDS-RS) for feature selection and simplified swarm optimization for intrusion data classification. IDS-RS is proposed to select the most relevant features that can represent the pattern of the network traffic. In order to improve the performance of SSO classifier, a new weighted local search (WLS) strategy incorporated in SSO is proposed. The purpose of this new local search strategy is to discover the better solution from the neighborhood of the current solution produced by SSO. The performance of the proposed hybrid system on KDDCup 99 dataset has been evaluated by comparing it with the standard particle swarm optimization (PSO) and two other most popular benchmark classifiers. The testing results showed that the proposed hybrid system can achieve higher classification accuracy than others with 93.3% and it can be one of the competitive classifier for the intrusion detection system.

A Novel Image Classification Algorithm Using Swarm-Based Technique for Image Database

Image data has become one of the most popular data type distributed in many multimedia applications. The effectiveness of image deployment is greatly dependent on the ability to classify and retrieve the image files based on their properties or content. However, image classification has faced a problem where the number of possible different combination of variables is very high. The algorithms which based on exhaustive search are unable to cope with the problem as the computational ability become infeasible. In this paper, a new image classification algorithm namely Simplified Swarm Optimization (SSO) has been proposed. This new approach is capable to obtain the high quality potential solution in the population which contributes to the improvement of the classification performance. This algorithm has been tested using image dataset which consists of seven classes of outdoor images. Moreover, the performance of SSO, Particle Swarm Optimization (PSO) and Support Vector Machine (SVM) have been compared and analyzed. The testing results show that SSO is more competitive than PSO and SVM, and can be fruitfully exploited in image database and solving image classification problem.

Benchmark of feature selection techniques with machine learning algorithms for cancer datasets

Classification is a technique based on machine learning used to classify each item in a set of data into a set of predefined classes or group. It is widely used in medical field to classify the medical data. In producing better classification result, feature selection been applied in many of the classification work as part of preprocessing step, where a subset of feature been used rather than the whole features from particular dataset. Feature selection eliminates irrelevant attribute to obtain high quality features that may contribute in enhancing classification process and producing better classification results. This study is conducted with the intention to focus on feature selection techniques as a method that helps classifiers producing better classification performance with the most significant features. During the experiments, a comparison between benchmark feature selection methods based on three cancer datasets and four well recognized machine learning algorithms has been made. This paper then analyzes the performance of all classifiers with and without feature selection in term of ROC and F-Measure. The study found that although there are no single feature selection method can satisfy all datasets, the results still effectively support the fact that feature selection helps in increasing the classifier performance with existence of minimum number of features.

Functional Link Neural Network with Modified Cuckoo Search Training Algorithm for Physical Time Series Forecasting

Functional link neural network (FLNN) naturally extends the family of theoretical feedforward network structure by introducing nonlinearities in inputs patterns enhancements. It has emerged as an important tool used for function approximation, pattern recognition and time series prediction. The standard learning algorithm used for the training of FLNN is the Backpropagation (BP) learning algorithm. However, one of the crucial problems with BP algorithm is it tends to easily get trapped in local minima, resulting the degrading performance of FLNN. To overcome this problem, this work proposed an alternative learning scheme for FLNN by using a Modified Cuckoo Search algorithm (MCS), and the model is called FLNN-MCS. The performance of FLNN-MCS is evaluated based on the prediction error, testing on two physical time series data; relative humidity and temperature. Simulation results have shown that the prediction performed by FLNN-MCS is much superior compared to Multilayer Perceptron and FLNN trained with BP, and FLNN trained with Artificial Bee Colony algorithm. The significant performance has proven that FLNN-MCS is capable in mapping the input-output function for the next-day ahead forecasting.

A Performance Comparison of Genetic Algorithm’s Mutation Operators in n-Cities Open Loop Travelling Salesman Problem

Travelling Salesman Problem (TSP) is one of the most commonly studied optimization problem. In Open Loop Travelling Salesman Problem (OTSP), the salesman travels to all the given m cities but does not return to the city he started and each city is visited by salesman exactly once. However, a new problem of OTSP occur when the salesman does not visit all the given m cities, but only to visit n cities from the given m cities. This problem called n-Cities Open Loop Travelling Salesman Problem (nOTSP), which seems to be more close to the real-life transportation problem. In this paper, Genetic Algorithm (GA) with different mutation operators is implemented to the nOTSP in order to investigate which mutation operators give the optimal solution in minimizing the distance and computational time of the n visited cities. The mutation operators are inversion, displacement, pairwise swap and the combination of the above three operators. The results of these comparisons show that the GA-inversion mutation operator can achieve better solution in minimizing the total distance of the tour. In addition, the GA with combination of three mutation operators has great potential in reducing the computation time.

Improvement of Audio Feature Extraction Techniques in Traditional Indian Musical Instrument

Traditional Indian musical instrument is one of the oldest musical instruments in the world. The musical instruments have their own importance in the field of music. Traditional Indian musical instrument could be categorized into three types such as stringed instruments, percussion instruments and wind-blown instruments. However, this paper will focus on string instruments because its show fluctuating behavior due to noise. Therefore, three techniques are selected based on the frequently used by previous researches which show some shortcoming while extracting noisy signal. The three techniques are Mel-Frequency Cepstral Coefficient (MFCC), Linear Predictive Coding (LPC) and Zero-Crossing Rate (ZCR). Hence, this research attempts to improve the feature extracting techniques by integrating Zero Forcing Equalizer (ZFE) with those extraction techniques. Three classifiers that are k-Nearest Neighbor (kNN), Bayesian Network (BNs) and Support Vector Machine (SVM) are used to evaluate the performance of audio classification accuracy. The proposed technique shows better classification accuracy when dealing with noisy signal.

2D Platform-based Game of Arithmetic Game-based Learning

Mobile games have attracted attention among game developers and users. Throughout the years, mobile games have developed to become one of the domination of the digital world. They have evolved from just a non-coloured with few animated dots and lines to a coloured and more realistic three-dimensional gameplay. Besides, mobile games do not only act as an entertainment but also as a medium for improving mental arithmetic skill if the content is designed accordingly. However, most of mobile games nowadays have been developed for leisure purpose only. There is less game on IQ test and education while more on skills and action alternatively. Hence, arithmetic game-based learning application is developed as an alternative to be included in the variety of gameplay or content of the game. The game created based on platform genre with endless running style. Arithmetic questions are provided in the gameplay as one of the obstacles to both provide challenges and IQ test for the players. This game is implemented in mobile platform. Technology Acceptance Model (TAM) is adopted to measure the game quality based on the evaluation of user acceptance level towards the gameplay, functionality and playability, and overall performance of the game. Overall, 50% of respondents agreed that the gameplay is satisfying. In addition, 40% of respondents strongly agreed that the functionality and playability of the game is stable and the overall performance of the game functions perfectly.

Flatten-T Swish: a thresholded ReLU-Swish-like activation function for deep learning

Activation functions are essential for deep learning methods to learn and perform complex tasks such as image classification. Rectified Linear Unit (ReLU) has been widely used and become the default activation function across the deep learning community since 2012. Although ReLU has been popular, however, the hard zero property of the ReLU has heavily hindering the negative values from propagating through the network. Consequently, the deep neural network has not been benefited from the negative representations. In this work, an activation function called Flatten-T Swish (FTS) that leverage the benefit of the negative values is proposed. To verify its performance, this study evaluates FTS with ReLU and several recent activation functions. Each activation function is trained using MNIST dataset on five different deep fully connected neural networks (DFNNs) with depth vary from five to eight layers. For a fair evaluation, all DFNNs are using the same configuration settings. Based on the experimental results, FTS with a threshold value, T=-0.20 has the best overall performance. As compared with ReLU, FTS (T=-0.20) improves MNIST classification accuracy by 0.13%, 0.70%, 0.67%, 1.07% and 1.15% on wider 5 layers, slimmer 5 layers, 6 layers, 7 layers and 8 layers DFNNs respectively. Apart from this, the study also noticed that FTS converges twice as fast as ReLU. Although there are other existing activation functions are also evaluated, this study elects ReLU as the baseline activation function.

Improved Functional Link Neural Network Learning Using Modified Bee-Firefly Algorithm for Classification Task

Functional Link Neural Network (FLNN) has been becoming as an important tool used in many applications task particularly in solving a non-linear separable problems. This is due to its modest architecture which required less tunable weights for training as compared to the standard multilayer feed forward network. The most common learning scheme for training the FLNN is a Backpropagation (BP-learning) algorithm. However, learning method by BP-learning algorithm tend to easily get trapped in local minima especially when dealing with non-linearly separable classification problems which affect the performance of FLNN. This paper discussed the implementation of modified Artificial Bee Colony with Firefly algorithm for training the FLNN network to overcome the drawback of BP-learning scheme. The aim is to introduce an alternative learning scheme that can provide a better solution for training the FLNN network for classification task.

The effectiveness of Bat algorithm for data handling in various applications

Feature selection is a technique used to reduce irrelevant data and finding the most relevant features that would increase classification accuracy. It is widely used in various applications such as medical, agriculture and Information Technology. In producing better classification result, feature selection been applied in many of the classification works as part of preprocessing step; where only a subset of feature been used rather than the whole features from a particular dataset. This research is conducted with the intention to find the appropriate data types according to the percentage of attributes reduction and classification performance. During the experiments, the effectiveness of data handling for Bat algorithm is tested via type of data and size of attributes in generic dataset. 10 datasets from UCI repository from various applications are used. The selected features are selected using Bat algorithm and measured by three classifiers; k-Nearest Neighbor (kNN), Naïve Bayes (NB) and Decision Tree (DT). This paper then analyzes the performance of all classifiers with and without feature selection in term of accuracy, sensitivity, F-Measure and ROC. The research found that although the percentage of reduction is high, it produces lowest result in classification performance since the type of data and number of attribute are not appropriate.