IJISA Vol. 7, No. 6, May. 2015
Cover page and Table of Contents: PDF (size: 197KB)
Various types of online learning algorithms have been developed so far to handle concept drift in data streams. We perform more detailed evaluation of these algorithms through new performance metrics - prequential accuracy, kappa statistic, CPU evaluation time, model cost, and memory usage. Experimental evaluation using various artificial and real-world datasets prove that the various concept drifting algorithms provide highly accurate results in classifying new data instances even in a resource constrained environment, irrespective of size of dataset, type of drift or presence of noise in the dataset. We also present empirically the impact of various features- size of ensemble, period value, threshold value, multiplicative factor and the presence of noise on all the key performance metrics.[...] Read more.
Neural network models are powerful tools for extracting the underlying dependency of a set of input/output data. However, the mentioned tools are in danger of sticking in local minima. The present study went to step forward by optimizing neural network models using three intelligent optimization algorithms, including genetic algorithm (GA), particle swarm optimization (PSO), and ant colony (AC), to eliminate the risk of being exposed to local minima. This strategy was capable of significantly improving the accuracy of a neural network by optimizing network parameters such as weights and biases. Nuclear magnetic resonance (NMR) log measures some of the most useful characteristics of reservoir rock; the capabilities of the optimized models were used for prediction of nuclear magnetic resonance (NMR) log parameters in a carbonate reservoir rock of Iran. Conventional porosity logs, which are the easily accessible tools compared to NMR log’s parameters, were introduced to the models as inputs while free fluid porosity and permeability, which were measured by NMR log, are desire outputs. The performance of three optimized models was verified by some unseen test data. The results show that PSO-based network and ACO-based network is the best and poorest method, respectively, in terms of accuracy; however, the convergence time of GA-based model is considerably smaller than PSO-based and GA-based models.[...] Read more.
Efficient feature selection is an important phase of designing an effective text categorization system. Various feature selection methods have been proposed for selecting dissimilar feature sets. It is often essential to evaluate that which method is more effective for a given task and what size of feature set is an effective model selection choice. Aim of this paper is to answer these questions for designing Urdu text categorization system. Five widely used feature selection methods were examined using six well-known classification algorithms: naive Bays (NB), k-nearest neighbor (KNN), support vector machines (SVM) with linear, polynomial and radial basis kernels and decision tree (i.e. J48). The study was conducted over two test collections: EMILLE collection and a naive collection. We have observed that three feature selection methods i.e. information gain, Chi statistics, and symmetrical uncertain, have performed uniformly in most of the cases if not all. Moreover, we have found that no single feature selection method is best for all classifiers. While gain ratio out-performed others for naive Bays and J48, information gain has shown top performance for KNN and SVM with polynomial and radial basis kernels. Overall, linear SVM with any of feature selection methods including information gain, Chi statistics or symmetric uncertain methods is turned-out to be first choice across other combinations of classifiers and feature selection methods on moderate size naive collection. On the other hand, naive Bays with any of feature selection method have shown its advantage for a small sized EMILLE corpus.[...] Read more.
Parkinson is a neurological disease and occurs due to lack of dopamine neurons. These dopamine neurons manage all body movements. Parkinson patients have difficulty in doing all daily routine activities, and also have disturbed vocal fold movements. Using voice analysis disease can be diagnosed remotely at an early stage with more reliability and in an economic way. In this paper, we have used 23 features dataset, all the features are analyzed and 15 features are selected from the total dataset. As in Parkinson tremor is present in the voice box muscles, so the variation in the period and amplitude of consecutive vocal cycles is present. The feature dataset selected consist of jitter, shimmer, harmonic to noise ratio, DFA, spread1 and PPE. Various classifiers are used and their comparison is done to find out which classifier is perfect in this environment. It is concluded that support vector classifiers as the best one with an accuracy of 96%. In the neural network classifiers with different transfer functions, there is tradeoff among the performance parameters.[...] Read more.
In this paper, a new acquisition protocol is adopted for identifying individuals from electroencephalogram signals based on eye blinking waveforms. For this purpose, a database of 10 subjects is collected using Neurosky Mindwave headset. Then, the eye blinking signal is extracted from brain wave recordings and used for the identification task. The feature extraction stage includes fitting the extracted eye blinks to auto-regressive model. Two algorithms are implemented for auto-regressive modeling namely; Levinson-Durbin and Burg algorithms. Then, discriminant analysis is adopted for classification scheme. Linear and quadratic discriminant functions are tested and compared in this paper. Using Burg algorithm with linear discriminant analysis, the proposed system can identify subjects with best accuracy of 99.8%. The obtained results in this paper confirm that eye blinking waveform carries discriminant information and is therefore appropriate as a basis for person identification methods.[...] Read more.
The Ad Hoc On-Demand Multi-Path Distance Vector (AOMDV) routing protocol allows the transport of data along one or more paths in wireless sensor networks (WSNs). The path chosen is based on a single shortest path hop count metric. The data on some WSNs is mission critical, for example, military and health care applications. Hence, fault tolerance in WSNs is becoming increasingly important. To improve the fault tolerance of WSNs in lossy environments, this work adds to the AOMDV routing protocol as it incorporates an additional packet loss metric. This Multi-criteria AOMDV or M-AOMDV is evaluated using the ns2 simulator. Simulations show that M-AOMDV maintains relatively low packet loss rates when the WSN is experiencing loss.[...] Read more.
Energy saving and improve the life time of the sensor node is main focus in the recent years for the researchers hence one of the application domain (Street light monitoring and controlling) of sensor required attention towards this direction. For contributing in this domain we have proposed a scheme for Street light controlling using distributed topology control (TC). The optimize version of A3 protocol reduces the number of messages send/received by the sensors which ultimately leads to the reduction of energy requirement. Experiments are carried on street light scenario for different no. of nodes by maintaining communication using Zigbee protocol. The performance of our extension is evaluated using, no. of messages send/receive & energy consumed during topology building and our approach is having good results as compared to the approach used for this type of network.[...] Read more.
Several Text Categorization (TC) techniques and algorithms have been investigated in the limited research literature of Arabic TC. In this research, Logistic Regression (LR) is investigated in Arabic TC. To the best of our knowledge, LR was never used for Arabic TC before. Experiments are conducted on Aljazeera Arabic News (Alj-News) dataset. Arabic text-preprocessing takes place on this dataset to handle the special nature of Arabic text. Experimental results of this research prove that the LR classifier is a competitive Arabic TC algorithm to the state of the art ones in this field; it has recorded a precision of 96.5% on one category and above 90% for 3 categories out of the five categories of Alj-News dataset. Regarding the overall performance, LR has recorded a macroaverage precision of 87%, recall of 86.33% and F-measure of 86.5%.[...] Read more.