Majlesi Journal of Multimedia Processing
Volume:4 Issue: 3, Sep 2015

By increase of fraud in the insurance company and imposition of immense financial costs on these companies, identification of, and prevention from, fraud has been turned into one of significant distresses of the insurance companies. Data mining techniques assist in discernment of, and prevention from, fraud and abuse in the insurance. In this paper, using two selection methods of feature based on genetic algorithm and correlation, influential variables were found and then implemented on the data set of naive bayesian and decision tree algorithms. Results showed that modeling by use of selection method of feature based on genetic algorithms has a higher accuracy compared to selection method of feature based on correlation.

With the technology advancement in recent years, many studies have been conducted on the diagnosis of heart disease and especially diagnosis of Pulmonary Artery Hypertension (PAH), since the clinical diagnosis of this disease is very difficult, using audio signal processing and spectral analysis of the fetal heart sound recordings, have given us an appropriate opportunity for PAH diagnosis. Our discussion in this study is the difference between heart sound recordings of children with PAH and those without PAH. This is the goal of our study in fact. In order to prove this theory and perform the study, we used number of infants with PAH less than 25 mm Hg (between 8 – 24 mm Hg) as healthy cases and several other with PAH greater or equal to 25 mm Hg (between 25 – 97 mm Hg), recording their heart sounds with an innovative and low cost method. It should be noted that for processing the resulted signals from recorded heart sounds, the non-linear cepstrum method were used and power spectral analysis were done in Matlab software. The obtained results in this study would suggest that the spectrum frequency range are close to the frequencies 15 – 25 Hz of the recorded heart sounds for those patients that their second left intercostal space were examined for getting the heart sound. The spectrum frequency range in people with PAH greater or equal to 25 mm Hg is considerably decreased in compare with those with PAH lower than 25 mm Hg. So the conclusion is that heart sound signals in PAH patients notably have a lower extent comparing with normal PAH people. The obtained data in frequency scope may be helpful in better diagnosis of PAH and also help the techniques and methods’ developments in auscultation the heart sounds for PAH diagnosis. In future, with regards to the available data and using them for diagnosis of such diseases, their analysis in time and frequency extent may be required.

Iris identification in compare with many biometric methods is highly accurate. The success of iris identification systems, are completely dependent on and limited to segmentation algorithm success. In this paper, an investigation into the Libor Masek identification [10] performance is considered in case that the segmentation algorithm is failed. Searching these cases allows finding the features which are essential. As well as, a common point related to Hamming distance threshold is calculated, regardless of segmentation performance success. Simulation results show that complete iris images are necessary in order to having a highly secured recognition system, however, the proposed method with incomplete iris images was %96.5 successful in identification.

Recommender systems which are a subset of web mining are currently one of the widely applied aspects of data mining. Recommender systems help users more easily and quickly find products that they truly prefer amidst the enormous volume of information available to them. Since providing a user-friendly environment is one of the most important things in e-commerce, this branch of web mining is popular among researchers. In this paper we propose a method that combines personality traits into the traditional rating-based similarity computation in the framework of item-based recommender systems with the motivation to make good recommendations for new users who have rated a few items. We further compare our method with pure traditional ratings-based similarity and other similar systems in several experimental conditions. Experimental results shows that the proposed algorithm provides more advantages in terms of improving recommendation quality and it can efficiently address the new user problem.

Co-channel speech signal is generated when two or more people are talking at the same time over the same channel. Segments of co-channel speech that are still usable for speech processing algorithms are known as “usable” speech. Usable speech is a context-dependent concept, so we consider usable speech extraction in speaker identification application. So far many algorithms are proposed for usable speech extraction [6-8]. The common feature of all existing algorithms is that they exploit the periodical property of the usable frames, in time domain or in frequency domain. In this paper, we propose a discrete wavelet transform (DWT) based algorithm to exploit the periodical property of speech frames. The proposed algorithm is used in two different cases: first, in the voiced detection algorithm and second, in the usable speech extraction algorithm. To evaluate the proposed algorithm, the simulation results are presented in the paper. The simulation results show that the proposed algorithm has 3% improvement in usable speech detection compared with other algorithms.

Different ways of data classification have been presented. Of these, different neural networks can be used, which have been highly regarded because of their proper design and the least error, if properly designed; Also when optimization process is done by heuristic algorithms Also when optimization process is done by heuristic algorithms, the effect of slow convergence and been trapped at local optimum is enhanced than other training methods such as back propagation. In this paper, a MLP neural network is trained with continuous ACO to find the hyperplanes for classifying three reference datasets. The results are shown in comparison with the back propagation algorithm. Also the mean square error is calculated to show the accuracy of the proposed method