Joarder Kamruzzaman

Support Vector Machines and Other Pattern Recognition Approaches to the Diagnosis of Cerebral Palsy Gait

Accurate identification of cerebral palsy (CP) gait is important for diagnosis as well as for proper evaluation of the treatment outcomes. This paper explores the use of support vector machines (SVM) for automated detection and classification of children with CP using two basic temporal-spatial gait parameters (stride length and cadence) as input features. Application of the SVM method to a childrens dataset (68 normal healthy and 88 with spastic diplegia form of CP) and testing on tenfold cross-validation scheme demonstrated that an SVM classifier was able to classify the children groups with an overall accuracy of 83.33% [sensitivity 82.95%, specificity 83.82%, area under the receiver operating curve (AUC-ROC=0.88)]. Classification accuracy improved significantly when the gait parameters were normalized by the individual leg length and age, leading to an overall accuracy of 96.80% (sensitivity 94.32%, specificity 100%, AUC-DROC area=0.9924). This accuracy result was, respectively, 3.21% and 1.93% higher when compared to an linear discriminant analysis and an multilayer-perceptron-based classifier. SVM classifier also attains considerably higher ROC area than the other two classifiers. Among the four SVM kernel functions (linear, polynomial, radial basis, and analysis of variance spline) studied, the polynomial and radial basis kernel performed comparably and outperformed the others. Classifiers performance as functions of regularization and kernel parameters was also investigated. The enhanced classification accuracy of the SVM using only two easily obtainable basic gait parameters makes it attractive for identifying CP children as well as for evaluating the effectiveness of various treatment methods and rehabilitation techniques

Forecasting of currency exchange rates using ANN: a case study

In todays global economy, accuracy in forecasting the foreign exchange rate or at least predicting the trend correctly is of crucial importance for any future investment. The use of computational intelligence based techniques for forecasting has been proved extremely successful in recent times. In this paper, we developed and investigated three artificial neural network (ANN) based forecasting model using standard backpropagation (SBP), scaled conjugate gradient (SCG) and backpropagation with Baysian regularization (BPR) for Australian foreign exchange to predict six different currencies against Australian dollar. Five moving average technical indicators are used to build the models. These models were evaluated on five performance metrics and a comparison was made with traditional ARIMA model. All the ANN based models outperform ARIMA model. It is found that SCG based model performs best when measured on the two most commonly used metrics and shows competitive results when compared with BPR based model on other three metrics. Experimental results demonstrate that ANN based model can closely forecast the forex market.

z-SVM: an SVM for improved classification of imbalanced data

Recent literature has revealed that the decision boundary of a Support Vector Machine (SVM) classifier skews towards the minority class for imbalanced data, resulting in high misclassification rate for minority samples. In this paper, we present a novel strategy for SVM in class imbalanced scenario. In particular, we focus on orienting the trained decision boundary of SVM so that a good margin between the decision boundary and each of the classes is maintained, and also classification performance is improved for imbalanced data. In contrast to existing strategies that introduce additional parameters, the values of which are determined through empirical search involving multiple SVM training, our strategy corrects the skew of the learned SVM model automatically irrespective of the choice of learning parameters without multiple SVM training. We compare our strategy with SVM and SMOTE, a widely accepted strategy for imbalanced data, applied to SVM on five well known imbalanced datasets. Our strategy demonstrates improved classification performance for imbalanced data and is less sensitive to the selection of SVM learning parameters.

Artificial Neural Networks in Finance and Manufacturing

A Sample of Contents: Artificial Neural Networks: Applications in Finance and Manufacturing Hybrid-learning Methods for Stock-index Modeling.

EVOLUTIONARY OPTIMIZATION (EvOpt): A BRIEF REVIEW AND ANALYSIS

Evolutionary Computation (EC) has attracted increasing attention in recent years, as powerful computational techniques, for solving many complex real-world problems. The Operations Research (OR)/Optimization community is divided on the acceptability of these techniques. One group accepts these techniques as potential heuristics for solving complex problems and the other rejects them on the basis of their weak mathematical foundations. In this paper, we discuss the reasons for using EC in optimization. A brief review of Evolutionary Algorithms (EAs) and their applications is provided. We also investigate the use of EAs for solving a two-stage transportation problem by designing a new algorithm. The computational results are analyzed and compared with conventional optimization techniques.

Energy-Balanced Transmission Policies for Wireless Sensor Networks

Transmission policy, in addition to topology control, routing, and MAC protocols, can play a vital role in extending network lifetime. Existing transmission policies, however, cause an extremely unbalanced energy usage that contributes to early demise of some sensors reducing overall networks lifetime drastically. Considering cocentric rings around the sink, we decompose the transmission distance of traditional multihop scheme into two parts: ring thickness and hop size, analyze the traffic and energy usage distribution among sensors and determine how energy usage varies and critical ring shifts with hop size. Based on above observations, we propose a transmission scheme and determine the optimal ring thickness and hop size by formulating network lifetime as an optimization problem. Numerical results show substantial improvements in terms of network lifetime and energy usage distribution over existing policies. Two other variations of this policy are also presented by redefining the optimization problem considering: 1) concomitant hop size variation by sensors over lifetime along with optimal duty cycles, and 2) a distinct set of hop sizes for sensors in each ring. Both variations bring increasingly uniform energy usage with lower critical energy and further improves lifetime. A heuristic for distributed implementation of each policy is also presented.

SVM based models for predicting foreign currency exchange rates

Support vector machine (SVM) has appeared as a powerful tool for forecasting forex market and demonstrated better performance over other methods, e.g., neural network or ARIMA based model. SVM-based forecasting model necessitates the selection of appropriate kernel function and values of free parameters: regularization parameter and /spl epsiv/-insensitive loss function. We investigate the effect of different kernel functions, namely, linear, polynomial, radial basis and spline on prediction error measured by several widely used performance metrics. The effect of regularization parameter is also studied. The prediction of six different foreign currency exchange rates against Australian dollar has been performed and analyzed. Some interesting results are presented.In an erroneous footfall ground-reaction force-time recording, which may occur for people with disabilities or frail elderly individuals, the stance time (ST) can be either corrupted or missing. Previous methods to estimate missing ST require force-time data from multiple force platforms and are affected by inter-step variability. This paper presents a model based on support vector machine (SVM) that is capable of estimating the missing ST from the available vertical force-timing characteristics with significantly high accuracy. The model was built using features taken from a data set of 466 sample trials of 27 subjects. A test on 40 sample trials drawn from all the subjects revealed an average prediction accuracy of 96.63% (/spl plusmn/2.89%). In one-fourth of the test trials, the prediction error was within 1.0%. The model achieves considerable improvement over an artificial neural network based model built and tested on the same data set. The effect of kernel junction parameters and /spl epsiv/-insensitive loss function on prediction error is also analysed and presented.

Inchoate Fault Detection Framework: Adaptive Selection of Wavelet Nodes and Cumulant Orders

Inchoate fault detection for machine health monitoring (MHM) demands high level of fault classification accuracy under poor signal-to-noise ratio (SNR) which persists in most industrial environment. Vibration signals are extensively used in signature matching for abnormality detection and diagnosis. In order to guarantee improved performance under poor SNR, feature extraction based on statistical parameters which are immune to Gaussian noise becomes inevitable. This paper proposes a novel framework for adaptive feature extraction based on higher order cumulants (HOCs) and wavelet transform (WT) (AFHCW) for MHM. Features extracted based on HOCs have the tendency to mitigate the impact of Gaussian noise. WT provides better time and frequency domain analysis for the nonstationary signals such as vibration in which spectral contents vary with respect to time. In AFHCW, stationary WT is used to ensure linear processing on the vibration data prior to feature extraction, and it helps in mitigating the impact of poor SNR. K-nearest neighbor classifier is used to categorize the type of the fault. Simulation studies show that the proposed scheme outperforms the existing techniques in terms of classification accuracy under poor SNR.

An environment-aware mobility model for wireless ad hoc network

Simulation is a cost effective, fast and flexible alternative to test-beds or practical deployment for evaluating the characteristics and potential of mobile ad hoc networks. Since environmental context and mobility have a great impact on the accuracy and efficacy of performance measurement, it is of paramount importance how closely the mobility of a node resembles its movement pattern in a real-world scenario. The existing mobility models mostly assume either free space for deployment and random node movement or the movement pattern does not emulate real-world situation properly in the presence of obstacles because of their generation of restricted paths. This demands for the development of a node movement pattern with accurately representing any obstacle and existing path in a complex and realistic deployment scenario. In this paper, we propose a general mobility model capable of creating a more realistic node movement pattern by exploiting the concept of flexible positioning of anchors. Since the model places anchors depending upon the context of the environment through which nodes are guided to move towards the destination, it is capable of representing any terrain realistically. Furthermore, obstacles of arbitrary shapes with or without doorways and any existing pathways in full or part of the terrain can be incorporated which makes the simulation environment more realistic. A detailed computational complexity has been analyzed and the characteristics of the proposed mobility model in the presence of obstacles in a university campus map with and without signal attenuation are presented which illustrates its significant impact on performance evaluation of wireless ad hoc networks.

A hybrid of multiobjective Evolutionary Algorithm and HMM-Fuzzy model for time series prediction

In this paper, we introduce a new hybrid of Hidden Markov Model (HMM), Fuzzy Logic and multiobjective Evolutionary Algorithm (EA) for building a fuzzy model to predict non-linear time series data. In this hybrid approach, the HMMs log-likelihood score for each data pattern is used to rank the data and fuzzy rules are generated using the ranked data. We use multiobjective EA to find a range of trade-off solutions between the number of fuzzy rules and the prediction accuracy. The model is tested on a number of benchmark and more recent financial time series data. The experimental results clearly demonstrate that our model is able to generate a reduced number of fuzzy rules with similar (and in some cases better) performance compared with typical data driven fuzzy models reported in the literature.