The challenge of fraud detection, especially in credit card transactions, continues to grow as fraudsters adapt to increasingly sophisticated tactics. Traditional methods, including rule-based systems, are often limited by high false positives and poor adaptability to new fraud patterns. This study introduces a cutting-edge approach using an attention-based ensemble framework that synergizes the strengths of convolutional neural networks (CNNs), graph neural networks (GNNs), recurrent neural networks (RNNs), and long short-term memory (LSTM) models. A key innovation of this model is its confidence-driven gating mechanism, which dynamically combines the most reliable prediction, ensuring both accuracy and transparency. By incorporating dependent ordered weighted averaging (DOWA) and induced ordered weighted averaging (IOWA) operators, the model effectively integrates outputs from various classifiers, improving robustness. Additionally, SHAP-based feature selection reveals the most influential variables, providing deeper interpretability. Extensive evaluations across three diverse benchmark datasets demonstrate the superior performance of this framework, surpassing individual classifiers and existing ensemble models in both balanced and highly imbalanced settings. This research paves the way for more reliable and transparent fraud detection systems, capable of adapting to evolving fraud tactics in real-world financial environments.

Conventionally, deep odometry networks use objective functions that only penalize short-term deviations from the true path. Since such an objective does not impose any constraints on the long-term deviations from the path, a second consistency-based loss term may be added to lower long-term drift. However, maintaining a balance between the two loss terms is challenging and often treated as a design hyperparameter. To mitigate this balancing issue, we propose to use the uncertainty over both odometry and the long-term transformations in a maximum likelihood setting and allow the network to tune the weighting between the two loss terms. To this end, we derive the odometry uncertainty alongside the pose outputs using the network itself and to derive the covariance matrix over the integrated transformation, we propose to propagate the odometry uncertainty through each iteration. This formulation provides an adaptive and statistically consistent method to weigh the incremental and integrated loss terms against each other, noting the increase in uncertainty as more steps are integrated over. We show that our approach to consistency-based losses allows the network to surpass the accuracy of the state-of-the-art visual odometry approaches. Then, the efficacy of the derived uncertainty as weighting medium is visualized and the performance benefits of uncertainty quantification are shown in a pose-graph based localization scenario.

Pedestrian detection is one of the significant tasks in Autonomous Vehicles (AVs). There are two kinds of networks which are widely used for pedestrian detection: single-stage networks and region-based networks. Single-stage networks, such as YOLO, solve the bounding box regression and classification problems simultaneously which makes them faster than region-based networks such as Faster R-CNN. Nonetheless, the main structure of YOLO is too complex and slow for the pedestrian detection task in AVs and cannot detect small pedestrians. Furthermore, unlike region-based networks where all features of the region containing a pedestrian is used in classification, in YOLO only the features of a cell in which the center of anchor box lies is used in classification. In this paper, these issues related to YOLO will be addressed such that it can be better used for pedestrian detection.

Abstract Studying dynamic behaviours of a transportation system requires the use of the system mathematical models as well as prediction of traffic flow in the system. Therefore, traffic flow prediction plays an important role in today's intelligent transportation systems. This article introduces a new approach to short‐term daily traffic flow prediction based on artificial neural networks. Among the family of neural networks, multi‐layer perceptron (MLP), radial basis function (RBF) neural network and wavenets have been selected as the three best candidates for performing traffic flow prediction. Moreover, back‐propagation (BP) has been adapted as the most efficient learning scheme in all the cases. It is shown that the coefficients produced by temporal signals improve the performance of the BP learning (BPL) algorithm. Temporal signals provide researchers with a new model of temporal difference BP learning algorithm (TDBPL). The capability and performance of TDBPL algorithm are examined by means of simulation in order to prove that the wavelet theory, with its multi‐resolution ability in comparison to RBF neural networks, is a suitable algorithm in traffic flow forecasting. It is also concluded that despite MLP applications, RBF neural networks do not provide negative forecasts. In addition, the local minimum problems are inevitable in MLP algorithms, while RBF neural networks and wavenet networks do not encounter them.

Acquiring new customers in any business is much more expensive than trying to keep the existing ones. As a result, many prediction algorithms have been proposed to detect churning customers. In this paper, the ordered weighted averaging (OWA) technique is brought to the attention of marketing researchers. We have applied OWA technique to improve the prediction accuracy of existing churn management systems. The decision lists of underlying prediction algorithms have been fused using OWA algorithm. Applied to the database of a telecommunication company, this method is found to significantly improve accuracy in predicting churn compared to the best existing result in the literature of the churn management. Our findings lead us to believe that using OWA technique could cause to increase profit for the companies.

Naïve-Bayes and k-NN classifiers are two machine learning approaches for text classification. Rocchio is the classic method for text classification in information retrival. Based on these three approaches and using classifier fusion methods, we propose a novel approach in text classification. Our approach is a supervised method, meaning that the list of categories should be defined and a set of training data should be provided for training the system. In this approach, documents are represented as vectors where each component is associated with a particular word. We propose voting methods and OWA operator and Decision Template method for combining classifiers. Experimental results show that these methods decrese the classification error 15 percent as measured on 2000 training data from 20 newsgroups dataset.

Nowadays, the need for intelligent software systems as personal meta search engines which are capable of supplying user by needed information from massive information resources is sensible. Moreover, the current tools have many deficiencies. In a meta search engine, proper queries based upon user's interests are sent to different search engines (or in general, information servers). Then, the returned results are refined and made available for the user based on their priorities. On the same direction in this study we try to design, implement and examine a complete architecture for a customized software intelligent agent which is able to retrieve information from multiple sources based on user interests. The designed architecture is able to use information fusion methods to achieve this goal. Here, the OWA, which is a member of the fuzzy integral operators family, has been used. This agent performs fusion in data, feature, and decision levels. Also the agent is able to extract the behavioral model of each information server against various subjective clusters to increase the intelligence of searching methods and obtain higher quality results in searching process.

Nowadays, there are considerable attentions to combined classifier. Recently, the focus has been shifting from practical heuristic solutions of combination methods to give a methodological way of design. In this study a co-evolutionary algorithm is presented for this purpose. The algorithm synthesizes an explicit classifier directly from bserved data produced by intelligently generated tests. The algorithm is composed of two co-evolving populations; one population evolves candidate classifiers. The second population evolves informative tests that either extract new information from the pattern or elicit desirable behavior from it The fitness of candidate classifiers is their ability to classify in response to all tests carried out so far; the fitness of candidate tests is their ability to make the classifiers disagree in their classifications. The generality of this modeling-evaluation algorithm is demonstrated by applying the chosen classifier of this algorithm to identify modulation methods and results depict the power of this algorithm.

Today due to the importance and necessity of implementing security systems in homes and buildings with the capability of higher certainty and lower cost, sensor fusion methods as applicable and high performance methods are attracting the researchers' attention. In this paper, the application of Dempster-Shafer evidential theory and the more general and newer one Dezert-Smarandache theory for implementing a home security system using sensor data fusion have been considered. The benefits of multisensor fusion in comparison with the traditional single sensor systems or systems in which any sensor itself connects to the control unit have been shown.

Errors due to sensor bias are often present in sensor data and can reduce the tracking accuracy and stability of multi-sensor systems. The other practical problem is that the target data reported by the sensors are usually not time-coincident or synchronous due to the different data. This paper deals with these problems and presents a new algorithm for estimation of both constant and dynamic biases in asynchronous multisensor systems. We use the measurements from asynchronous sensors into pseudo measurements of the sensor biases with additive noises that are zero-mean, white and with easily calculated covariances. This algorithm is a Kahnan filter based technique to estimate both the range and offset biases and is implemented recursively which is computationally efficient and provided real time estimation of asynchronous sensor bias. The Simulation results show the Cramer-Rao Lower Bound (CRLB) is achievable. This means the proposed estimation algorithm is statistically efficient.

In this paper, first an enhanced NeuroFuzzy method for modeling nonlinear system is presented. In this method we use EM algorithm for identification of local models, which gain us model mismatch covariance. The achieved model can be stated in state space model as a linear time-varying system. As the noise and model mismatch covariance is known, Kalman filter can be easily used for centralized estimation fusion. The simulations show that using centralized estimation fusion will enhance the estimation accuracy to a great deal.

This paper presents a novel methodology for long term prediction of chaotic time series based on spectral analysis and neuro fuzzy modeling. A main motivation of using spectral analysis is to find some long term predictable components which describe the time series dynamics properly. In addition, this paper proposes a novel input variables selection criterion which is based on correlation analysis. The objective of this algorithm is to maximize relevance between inputs and output and minimizes the redundancy of selected inputs. After selecting input variables, a locally linear neuro fuzzy model is optimized for each of the principal components obtained from singular spectrum analysis, and the multi step predicted values are recombined to make the natural chaotic phenomenon. Two case studies are considered in this paper. The method has been applied to the long-term prediction of disturbance storm time (DST) as a solar activity indexes and one time series from neural forecasting competitions NN3. Results depict the power of the proposed method in long-term prediction of chaotic time series.

Today there are thousands of search engines available, so it is difficult for users to know where they are, how to use them and what topics they best address. Meta-search engines reduce the users' burden by dispatching queries to multiple search engines in parallel and combining the returned results. But there are some problems yet. One of them is that, none of them includes the user model in the answer. In this paper, we propose a mechanism to create a mapping between different categories of users and the underlying search engines using the reinforcement learning approach. By this way, the meta-search engine learns to identify which search engines are most appropriate for particular queries from different user models.

The cyclic solar activity has significant effects on earth, climate, satellites and space missions. Several methods have been introduced for the prediction of sunspot number, which is a common measure of solar activity. In this study a co-evolutionary algorithm is presented for inferring the topology and parameters of a multilayered neural network with the minimum of experimentation to the sunspot number time series which will be used as a predictor in predicting such phenomena. The algorithm synthesizes an explicit model directly from the observed data produced by intelligently generated tests. The algorithm is composed of two co-evolving populations; one population evolves candidate neural networks. The second population evolves informative tests that either extract new information from the hidden system or elicit desirable behavior from it. The fitness of candidate neural networks is their ability to explain behavior of the target chaotic system observed in response to all tests carried out so far; the fitness of candidate tests is their ability to make the models disagree in their predictions. The generality of this modeling-evaluation algorithm is demonstrated by applying the chosen model of this algorithm to predict sunspot number and the results depict the power of this training method which yields proper model to predict such chaotic time series.

In this paper we propose a procedure for recursive identification of discrete time piecewise affine (PWA) hybrid systems. The PWA identification problem involves the estimation of both the parameters of affine submodels and the partition of the PWA map from data. In this paper the submodel parameters estimation problem is solved via recursive k-plane clustering algorithm and the problem of region estimation is performed by incremental proximal support vector machine. Also, the effect of mode switching on the estimated parameters convergence is investigated

Several contemporary commercial search engines and meta-search engines attempt to produce more relevant results for a particular query intelligently. In this paper, a novel meta-search engine, named as WebFusion, has been introduced. This meta-search engine learns the expertness of each underlying search engine in a certain category based on the users' preferences. Moreover, an intelligent re-ranking method is proposed based on OWA. This re-ranking method is used to fuse the results' scores of the underlying search engines. WebFusion uses the click-through data concept to give a content-oriented ranking score to each result page. Click-through data concept is the implicit feedback of the users' preferences, which is also used as a reinforcement signal in the learning process, to predict the users' preferences and reduces the seeking time in the search result list. This research provides a direct mapping between users' categories and the underlying search engines, based on users' judgments. Experimental results showed that the average click rate and the variance of clicks are noticeably decreased comparing with ProFusion while the relevancy of the responses is increased.

The paper presents the design of fuzzy sliding mode control based on genetic algorithms. An optimal design criterion is not required in the design of a fuzzy sliding mode control. Therefore, we design a fuzzy sliding mode controller for the general nonlinear control systems as an optimization problem and apply the optimal searching algorithms and genetic algorithms to find the optimal rules and membership functions of the controller. The proposed approach has the merit to determine the optimal structure and the inference rules of fuzzy siding mode controller simultaneously. Using the proposed approach, the tracking problem of two-degree-of-freedom rigid robot manipulator is studied. Simulation results of the close-loop system with the proposed controller based on genetic algorithms show the effectiveness of the approach.