Adaptive anti-spam filtering for agglutinative languages: a special case for Turkish

We propose anti-spam filtering methods for agglutinative languages in general and for Turkish in particular. The methods are dynamic and are based on Artificial Neural Networks (ANN) and Bayesian Networks. The developed algorithms are user-specific and adapt themselves with the characteristics of the incoming e-mails. The algorithms have two main components. The first one deals with the morphology of the words and the second one classifies the e-mails by using the roots of the words extracted by the morphological analysis. Two ANN structures, single layer perceptron and multi-layer perceptron, are considered and the inputs to the networks are determined using binary model and probabilistic model. Similarly, for Bayesian classification, three different approaches are employed: binary model, probabilistic model, and advanced probabilistic model. In the experiments, a total of 750 e-mails (410 spam and 340 normal) were used and a success rate of about 90% was achieved.     

Knowledge discovery using neural approach for SME’s credit risk analysis problem in Turkey

This study proposes a knowledge discovery method that uses multilayer perceptron (MLP) based neural rule extraction (NRE) approach for credit risk analysis (CRA) of real-life small and medium enterprises (SMEs) in Turkey. A feature selection and extraction stage is followed by neural classification that produces accurate rule sets. In the first stage, the feature selection is achieved by decision tree (DT), recursive feature extraction with support vector machines (RFE-SVM) methods and the feature extraction is performed by factor analysis (FA), principal component analysis (PCA) methods. It is observed that the RFE-SVM approach gave the best result in terms of classification accuracy and minimal input dimension. Among various classifiers k-NN, MLP and SVM are compared in classification experiments. Then, the Continuous/Discrete Rule Extractor via Decision Tree Induction (CRED) algorithm is used to extract rules from the hidden units of a MLP for knowledge discovery. Here, the MLP makes a decision for customers as being “good” or “bad” and reveals the rules obtained at the final decision. In the experiments, Turkish SME database has 512 samples. The proposed approach validates the claim that is a viable alternative to other methods for knowledge discovery.     

Advanced Oxidation of Diuron by Photo-Fenton Treatment as a Function of Operating Parameters

Advanced oxidation of diuron in aqueous solution by photo-Fenton treatment was investigated by batch experiments. Effects of operating parameters, namely, the concentrations of pesticide (diuron), hydrogen peroxide (H2O2), and ferrous ion [Fe (II)] on oxidation of diuron were investigated by using Box-Behnken statistical experiment design and the response surface methodology. Diuron oxidation by photo-Fenton treatment was evaluated by determining the total organic carbon (TOC), diuron, and adsorbable organic halogen (AOX) removals. Concentration ranges of the reagents resulting in the highest level of diuron oxidation were determined. Diuron removal increased with increasing H2O2 and Fe (II) concentrations, up to a certain level. H2O2 concentration had a more profound effect than diuron and Fe (II) on removal of diuron, TOC, and AOX from the aqueous solution. Complete (100%) disappearance of diuron was achieved after a 15?min reaction period. However, 85% of diuron was mineralized after 240?min, indicating a low level of intermediate formation. Optimal H2O2/Fe (II)/diuron ratio resulting in maximum diuron (100%), TOC (85%), and AOX (100%) removals was found to be 267/36/25?(mg?L?1).     

Aircraft Sensor and Actuator Fault Detection, Isolation, and Accommodation

The purpose of the paper is to present an approach to detect, isolate, and accommodate the aircraft sensor and actuator faults using unknown input observers (UIOs). Full-order observers, reduced-order observers, and UIOs are widely used in state estimations. After the estimation of states, fault detection can be provided by conducting residual analysis. Despite of the existence of unknown inputs, fault detection and isolation are implemented for a very large, four-engined, cargo jet aircraft model. Sensor accommodation is realized via switching under redundant sensor existence assumption. Actuator accommodation is provided by gain scheduling. Hence, if a fault occurs in an actuator corresponding to the control surfaces, the remainder (n?1) actuators are used to avoid hazardous flight. Sensor and actuator faults are detected by using residuals. Sensor faults are effective on the outputs, while actuator faults are effective on the state equations. Fault isolation is implemented by taking into account that each residual is sensitive to all of the other faults but one fault. Fault detection, isolation, and a recommended accommodation are shown through the simulations to be functional.     

Intrauterine growth restriction (IUGR) risk decision based on support vector machines

This paper studies the risk of intrauterine compromise in the fetuses with intrauterine growth restriction (IUGR) using support vector machines (SVM). A structured and globally optimized SVM system may be preferable procedure in the identification of IUGR fetus at risk. The IUGR risk is estimated in two stages: In the first stage, noninvasive Doppler pulsatility index (PI) and resistance index (RI) of umbilical artery (UA), middle cerebral artery (MCA) and ductus venosus (DV), and amniotic fluid index (AFI) are retrospectively analyzed and the Doppler indices are applied to the SVM system to make a diagnosis decision on the fetal well being as “reactive” or “nonreactive and/or fetal distress (FD)” on the nonstress test (NST) (training data). In the second stage (testing data), the decision is validated by the NST (target value). Experiments are performed in retrospective clinical situation. Forty-four preterm with IUGR and without IUGR pregnancies before 34 weeks gestation are considered. Also, the nonparametric Bayes-risk decision rule, k-nearest neighbor (k-NN), is used for comparison. It is observed that the SVM system is proven to be useful in predicting the expected risk in IUGR cases in the small population study. Also, the PI and RI values of UA, MCA and DV are effective in distinguishing IUGR cases at risk.     

Modeling and rheology of HTPB based composite solid propellants

The propellant with the minimum viscosity required for a defect-free casting can be obtained by proper selection of the size and fractions of solid components leading to maximum packing density. Furnas' model was used to predict the particulate composition for the maximum packing density. Components with certain size dispersions were combined to yield a size distribution that is closest to the optimum one given by Furnas for maximum packing. The closeness of the calculated size distribution to the optimum one was tested by using the least square technique. The results obtained were experimentally confirmed by viscosity measurement of uncured propellants having HTPB binder and trimodal solid part accordingly prepared by using aluminum (volumetric mean particle diameter of 10.4 μm) and ammonium perchlorate with four different sizes (volumetric mean particle diameters: 9.22, 31.4, 171, and 323 μm). The experimental measurements showed that the compositions for the minimum viscosity are in good agreement with those predicted by using the model for maximum packing. The propellant consisting of particles with mean diameters of 10.4, 31.4, and 323 μm was found to yield the minimum viscosity. This minimum viscosity was observed when the fraction of the sizes with respect to total solids was 0.141, 0.300, and 0.559, respectively.     

Fine‐tuning the mechanical properties of hydroxyl‐terminated polybutadiene/ammonium perchlorate‐based composite solid propellants by varying the NCO/OH and triol/diol ratios

Changes in the mechanical properties of hydroxyl‐terminated polybutadiene/ammonium perchlorate‐based composite solid propellants were studied during the curing period with respect to variations in the crosslink density, which was predominantly determined by the equivalent ratio of diisocyanate to total hydroxyl (NCO/OH ratio) and the equivalent ratio of triol to diol (triol/diol ratio). For this purpose, 16 propellants were prepared in different compositions through changes in the NCO/OH ratios (0.81, 0.82, 0.83, and 0.85) for each triol/diol ratio (0.07, 0.09, 0.11, and 0.13) and were tested for their mechanical properties immediately after curing. The propellants with an NCO/OH ratio of 0.82 had minimum stress, modulus, and hardness with maximum strain capability, whereas the propellants with an NCO/OH ratio of 0.85 showed just the opposite behavior. Variations in the isocyanate level seemed to have more effect on the mechanical properties at higher triol/diol ratios. It was also concluded that the propellants with triol/diol–NCO/OH combinations of 0.11–0.83, 0.11–0.85, 0.13–0.81, 0.13–0.83, and 0.13–0.85 were not acceptable for upper stage case‐bonded rocket applications because of either high tensile strength or high modulus. © 2002 Wiley Periodicals, Inc. J Appl Polym Sci 84: 2072–2079, 2002; DOI 10.1002/app.10605     

The Effect of Calcined Colemanite Addition on the Mechanical Strength of Magnetite Pellets Produced with Organic Binders

Iron ore pellets must have sufficient mechanical strengths against degradation in all stages of pellet production. Low strength is also a problem for product pellets since they abrade during transportation to the reduction furnaces. The use of a binder is necessary to provide sufficient strength to the pellets and for better operation and handling of pellets. Bentonite is the standard binder in the industry; however, it is considered an impurity due to its acid oxide contents. Organic binders have been tested for many years as alternative binder to bentonite. They have been found to give sufficient wet pellet properties. However, they failed to provide sufficient strength to the preheated and fired pellets due to lack of slag bonding. It has been assumed that one possible effective method to improve the preheated and fired pellet strengths is addition of a slag-bonding constituent. In this study, calcined colemanite was added to the pellet feed to overcome the lower strength problem encountered with organic binder use. The strength of pellets produced with organic binders and calcined colemanite alone and in combination was comparatively studied against the strength of pellets made with standard bentonite binder in magnetite concentrate pelletizing. The results showed that addition of calcined colemanite into the pellet mixture improved the preheated and fired pellet strengths of pellets produced with organic binders.     

Performance of local search heuristics on scheduling a class of pipelined multiprocessor tasks

This paper presents the evaluation of the solution quality of heuristic algorithms developed for scheduling multiprocessor tasks for a class of multiprocessor architectures designed to exploit temporal and spatial parallelism simultaneously. More specifically, we deal with multi-level or partitionable architectures where MIMD parallelism and multiprogramming support are the two main characteristics of the system. We investigate scheduling a number of pipelined multiprocessor tasks with arbitrary processing times and arbitrary processor requirements in this system. The scheduling problem consists of two interrelated sub-problems, which are finding a sequence of pipelined multiprocessor tasks on a processor and finding a proper mapping of tasks to the processors that are already being sequenced. For the solution of the second problem, various techniques are available. However, the problem remains of generating a feasible sequence for the pipelined operations. We employed three well-known local search heuristic algorithms that are known to be robust methods applicable to various optimization problems. These are Simulated Annealing, Tabu Search, and Genetic Algorithms. We then conduct computational experiments and evaluate the reduction achieved in completion time by each heuristic. We have also compared the results with well-known simple list-based heuristics.     

Algorithm of J-spectral factorization of polynomial matrices

Numerical algorithms for the canonical J-factorization of polynomial matrices with respect to the imaginary axis are given. The factorization problems for the non-regular polynomial matrices and for polynomials, whose determinants have zero roots, are considered. The constructed algorithms are not connected to the finding of roots of the polynomial. The basic calculation procedure is the construction of the stabilizing solution of the matrix algebraic Riccati equation.