Multiple Classifier Fusion Research Articles

Deep multiple classifier fusion for traffic scene recognition

The recognition of the traffic scene in still images is an important yet difficult task in an intelligent transportation systems. The main difficulty lies in how to improve the image processing algorithms for different traffic participants and the various layouts of roads while discriminating the different traffic scenes. In this paper, we attempt to solve the traffic scene recognition problem with three distinct contributions. First, we propose a deep multi-classifier fusion method in the setting of granular computing. Specifically, the deep multi-classifier fusion method which involves local deep-learned feature extraction at one end that is connected to the other end for classification through a multi-classifier fusion approach. At the local deep-learned feature extraction end, the operation of convolution to extract feature maps from the local patches of an image is essentially a form of information granulation, whereas the fusion of classifiers at the classification end is essentially a form of organization. The second contribution is the creation of new traffic scene data set, named the “WZ-traffic”. The WZ-traffic data set consists of 6035 labeled images, which belong to 20 categories collected from both an image search engine as well as from personal photographs. Third, we make extensive comparisons with state-of-the-art methods on the WZ-traffic and FM2 data sets. The experiment results demonstrate that our method dramatically improves traffic scene recognition and brings potential benefits to many other real-world applications.

Multiple classifiers fusion and CNN feature extraction for handwritten digits recognition

Handwritten digits recognition has been treated as a multi-class classification problem in the machine learning context, where each of the ten digits (0–9) is viewed as a class and the machine learning task is essentially to train a classifier that can effectively discriminate the ten classes. In practice, it is very usual that the performance of a single classifier trained using a standard learning algorithm is varied on different datasets, which indicates that the same learning algorithm may train strong classifiers on some datasets but weak classifiers may be trained on other datasets. It is also possible that the same classifier shows different performance on different test sets, especially when considering the case that image instances can be highly diverse due to the different handwriting styles of different people on the same digits. To address the above issue, development of ensemble learning approaches have been very necessary to improve the overall performance and make the performance more stable on different datasets. In this paper, we propose a framework that involves CNN-based feature extraction from the MINST dataset and algebraic fusion of multiple classifiers trained on different feature sets, which are prepared through feature selection applied to the original feature set extracted using CNN. The experimental results show that the classifiers fusion can achieve the classification accuracy of ≥ 98%.

Belief-based system for fusing multiple classification results with local weights

The fusion of multiple classifiers is an effective way to improve classification performance. Classifiers trained on different datasets generally have different qualities on classification. Moreover, the classification results of different objects (patterns) by a common classifier may also show different reliabilities. So, we propose a system for fusing multiple classification results with local weights based on belief functions theory. For one classifier, the training dataset is divided into some clusters and each cluster is used to train a weight to represent the reliability of this classifier used for classifying objects in this cluster. Thus, each classifier has multiple different weights corresponding to the patterns in different clusters. The weights can be optimized by minimizing the sum of distances between the weighted fusion results and the truths in all clusters. For each classifier, the object to classify is first assigned to the closest cluster according to its attributes, and then its classification result will be discounted with the corresponding weight. Multiple discounted results are combined using Dempster’s rule. To reduce the errors, a soft decision-making rule is developed by modeling the partial imprecision. If a hard decision shows a high risk of error, this object will be committed to a set of possible classes. Such imprecision that can be clarified using other techniques is usually considered better than an error. So classification efficiency of imprecise decision is defined to be lower than that of a correct result but higher than that of an error. For the object to classify, the final decision is made via comparing the efficiency of hard decisions with that of imprecise decisions using patterns in the closest cluster. Finally, some real datasets are used in experimental applications to demonstrate the effectiveness of the proposed method by comparison with other related fusion methods.

Risk assessment for enterprise merger and acquisition via multiple classifier fusion

This paper aims to solve the problem of Risk assessment for enterprise merger and acquisition (M&A), which is an important problem in modern company management. Firstly, we design an index system to assess risks of enterprise M&A behavior, and six risks are considered: 1) Systemic risk, 2) Law risk, 3) Financial risk, 4) Intermediary risk, 5) Integrated risk, and 6) Information risk. Furthermore, 18 indexes are chosen to cover these six aspects. Secondly, we illustrate how to utilize the proposed risk assessment in the decision system for enterprise M&A risk assessment. We separate the M&A risk assessment process to three steps, that is, 1) Before M&A, and 2) In M&A, and 3) After M&A. Particularly, after the risk assessment process, there are three decisions for enterprise managers, that is, 1) implement the original M&A plan, 2) modify the original M&A plan, and 3) refuse it. Thirdly, we propose the multiple classifier fusion based risk assessment algorithm, which aims to effectively combine the six support vector machines. To relax the limitation of the SVM classifier, we introduce the fuzzy theory in the multiple classifier fusion algorithm, and the category label assignment is determined by utilizing a maximum membership rule. Finally, we conduct an experiment to make performance evaluation by constructing a dataset which includes the M&A data of 200 enterprises, among which 185 enterprises are used as training dataset and others are regarded as testing dataset. Using ROC curve, MAE and MAPE as evaluation criterions, performance of the proposed method is compared with single SVM scheme. Experimental results demonstrate that combining multiple the SVM classifiers together, accuracy of M&A risk assessment is greatly enhanced.

On the use of DAG-CNN architecture for age estimation with multi-stage features fusion

Accurate facial age estimation is quite challenging, since ageing process is dependent on gender, ethnicity, lifestyle and many other factors, therefore actual age and apparent age can be quite different. In this paper, we propose a new architecture of deep neural networks namely Directed Acyclic Graph Convolutional Neural Networks (DAG-CNNs) for age estimation which exploits multi-stage features from different layers of a CNN. Two instants of this system are constructed by adding multi-scale output connections to the underlying backbone from two well-known deep learning architectures, namely VGG-16 and GoogLeNet. DAG-CNNs not only fuse the feature extraction and classification stages of the age estimation into a single automated learning procedure, but also utilized multi-scale features and perform score-level fusion of multiple classifiers automatically. Fine-tuning such models helps to increase the performance and we show that even “off-the-shelf” multi-scale features perform quite well. Experiments on the publicly available Morph-II and FG-NET databases prove the effectiveness of our novel method.

Robust multimodal biometric system based on optimal score level fusion model

Multimodal biometric systems fuse information from multiple modalities to overcome limitations of individual classifiers. Score level fusion of multiple classifiers can effectively combine information from different modalities. However, most of the multimodal biometric systems are impaired by conflicting classifier scores under dynamic environment, which results in degradation in system's robustness and reliability. To address this, we propose a multimodal biometric system based on an optimal score level fusion model. The key idea of this work is to optimally integrate three complementary biometric traits namely iris, finger vein and fingerprint. For this, individual classifier performance is optimized using evolutionary Backtracking Search Optimization Algorithm (BSA). In addition, conflicting beliefs from individual classifiers are resolved using proportional conflict redistribution rules (PCR-6) to obtain a concurrent solution. The system exhibits optimal behaviour under dynamic environment through boosting or suppression of concurrent classifiers and resolving conflicts among discordant classifiers. The proposed biometric system is evaluated over chimeric multimodal datasets created from benchmark images. On an average, we achieve an accuracy of 98.43% and an EER of 1.57%. The proposed biometric system not only outperforms state-of-the-art techniques but also shows directions towards development of an expert multimodal biometric system.

Classifying impurity ranges in raw sugarcane using laser-induced breakdown spectroscopy (LIBS) and sum fusion across a tuning parameter window

The presence in raw sugarcane of low levels of solid impurities from soil particles and green and dry/brown sugarcane leaves is relevant to improving sugar mill production performance. Two ranges of impurities for raw sugar manufacturing processes need to be characterized from 0 to 5 wt% (desired material) and 8 to 10 wt% (undesired material); these ranges are denoted as 1 and 2, respectively. Laser-induced breakdown spectroscopy (LIBS) combined with chemometrics is used to detect chemical elements and different impurity ranges in leached raw sugarcane solutions. The potential use of LIBS based on leached solutions immobilized in a polyvinyl alcohol (PVA) polymer requires approximately 2 h sample preparation time. LIBS data are assigned to the above two impurity ranges using fusion of multiple classifiers. Most classifiers require a training set and optimization of a tuning parameter to select the best model; however, the sum fusion across a tuning parameter window used for classifying the samples in this study is a process that does not require either. The classification results are 97% accuracy for both ranges; 94% and 100% specificity for ranges 1 and 2, respectively; and 100% and 94% sensitivity for ranges 1 and 2, respectively. The classification results indicate potential for future applications in sugarcane refineries.

Multi-scale features for heartbeat classification using directed acyclic graph CNN

ABSTRACTA new architecture of deep neural networks, directed acyclic graph convolutional neural networks (DAG-CNNs), is used to classify heartbeats from electrocardiogram (ECG) signals into different subject-based classes. DAG-CNNs not only fuse the feature extraction and classification stages of the ECG classification into a single automated learning procedure, but also utilized multi-scale features and perform score-level fusion of multiple classifiers automatically. Therefore, DAG-CNN negates the necessity to extract hand-crafted features. In most of the current approaches, only the high level features which extracted by the last layer of CNN are used. Instead of performing feature level fusion manually and feeding the results into a classifier, the proposed multi-scale system can automatically learn different level of features, combine them and predict the output label. The results over the MIT-BIH arrhythmia benchmarks database demonstrate that the proposed system achieves a superior classification performance compared to most of the state-of-the-art methods.

Data fusion and multiple classifier systems for human activity detection and health monitoring: Review and open research directions

Activity detection and classification using different sensor modalities have emerged as revolutionary technology for real-time and autonomous monitoring in behaviour analysis, ambient assisted living, activity of daily living (ADL), elderly care, rehabilitations, entertainments and surveillance in smart home environments. Wearable devices, smart-phones and ambient environments devices are equipped with variety of sensors such as accelerometers, gyroscopes, magnetometer, heart rate, pressure and wearable camera for activity detection and monitoring. These sensors are pre-processed and different feature sets such as time domain, frequency domain, wavelet transform are extracted and transform using machine learning algorithm for human activity classification and monitoring. Recently, deep learning algorithms for automatic feature representation have also been proposed to lessen the burden of reliance on handcrafted features and to increase performance accuracy. Initially, one set of sensor data, features or classifiers were used for activity recognition applications. However, there are new trends on the implementation of fusion strategies to combine sensors data, features and classifiers to provide diversity, offer higher generalization, and tackle challenging issues. For instances, combination of inertial sensors provide mechanism to differentiate activity of similar patterns and accurate posture identification while other multimodal sensor data are used for energy expenditure estimations, object localizations in smart homes and health status monitoring. Hence, the focus of this review is to provide in-depth and comprehensive analysis of data fusion and multiple classifier systems techniques for human activity recognition with emphasis on mobile and wearable devices. First, data fusion methods and modalities were presented and also feature fusion, including deep learning fusion for human activity recognition were critically analysed, and their applications, strengths and issues were identified. Furthermore, the review presents different multiple classifier system design and fusion methods that were recently proposed in literature. Finally, open research problems that require further research and improvements are identified and discussed.

Adaptive multiple classifiers fusion for inertial sensor based human activity recognition

Aiming at the poor accuracy of single classifier in recognizing daily activities based on single accelerometer, this paper presents a method of daily activity recognition based on ensemble learning and full information matrix based fusion weight. Firstly, features from three attributes are extracted from the acceleration signals respectively. The three kinds of features can well describe the information of the activity, and they are relatively independent, which can reduce the interference caused by information redundancy in the process of fusion. Then three base classifiers of support vector machines are constructed based on three kinds of features respectively. Secondly, Euclidean distance between the test sample and every training sample for each type of feature vector is calculated to find out the k nearest neighbors of the test sample from the training set by the K-nearest neighbour method. The cluster analysis is used to compute the similarity between every neighbor and the test sample. Then, a proper threshold is utilized to remove the invalid neighbor whose similarity is less than the threshold. According to the effective neighbor, the full information matrix is constructed to calculate the accuracy. The weight of every single classifier is set dynamically according to the accuracy. Experiments showed that our proposed method get the best average recognition accuracy of 94.79% among several other weight functions when using majority voting method, besides, the time cost is also appealing.

Energy Efficiency Optimization Based on Context Awareness in Wearable Computing

The motion capture system of the wireless wearable device can achieve identification and classification of human motion through various classifiers, but wireless data transmission consumes most of the energy of the sensor node. An adaptive wireless transmission control algorithm based on context awareness is proposed realize energy efficiency optimization based on single classifier and multiple classifiers fusion. Firstly, adaptive wireless transmission control method based on fixed single classifier can effectively improve the overall energy efficiency under the limit of classification accuracy. Secondly, adaptive transmission control method of multi-classifier fusion is proposed with adjusting decision threshold based on various classifiers and their characteristics. The experimental results validate our proposed algorithm.

Multiple Classifier Fusion and Optimization for Automatic Focal Cortical Dysplasia Detection on Magnetic Resonance Images

In magnetic resonance (MR) images, detection of focal cortical dysplasia (FCD) lesion as a main pathological cue of epilepsy is challenging because of the variability in the presentation of FCD lesions. Existing algorithms appear to have sufficient sensitivity in detecting lesions but also generate large numbers of false-positive (FP) results. In this paper, we propose a multiple classifier fusion and optimization schemes to automatically detect FCD lesions in MR images with reduced FPs through constructing an objective function based on the F-score. Thus, the proposed scheme obtains an improved tradeoff between minimizing FPs and maximizing true positives. The optimization is achieved by incorporating the genetic algorithm into the work scheme. Hence, the contribution of weighting coefficients to different classifications can be effectively determined. The resultant optimized weightings are applied to fuse the classification results. A set of six typical FCD features and six corresponding Z-score maps are evaluated through the mean F-score from multiple classifiers for each feature. From the experimental results, the proposed scheme can automatically detect FCD lesions in 9 out of 10 patients while correctly classifying 31 healthy controls. The proposed scheme acquires a lower FP rate and a higher F-score in comparison with two state-of-the-art methods.

Using AdaBoost-based Multiple Functional Neural Fuzzy Classifiers Fusion for Classification Applications

In this study, two intelligent classifiers, the AdaBoost-based incremental functional neural fuzzy classifier (AIFNFC) and the AdaBoost-based fixed functional neural fuzzy classifier (AFFNFC), are proposed for solving the classification problems. The AIFNFC approach will increase the amount of functional neural fuzzy classifiers based on the corresponding error during the training phase; while the AFNFC approach is equipped with a fixed amount of functional neural fuzzy classifiers. Then, the weights of AdaBoost procedure are assigned for classifiers. The proposed methods are applied to different classification benchmarks. Results of this study demonstrate the effectiveness of the proposed AIFNFC and AFFNFC methods.

Improved Face Recognition Across Poses using Fusion of Probabilistic Latent Variable Models

Uncontrolled environments have often required face recognition systems to identify faces appearing in poses that are different from those of the enrolled samples. To address this problem, probabilistic latent variable models have been used to perform face recognition across poses. Although these models have demonstrated outstanding performance, it is not clear whether richer parameters always lead to performance improvement. This work investigates this issue by comparing performance of three probabilistic latent variable models, namely PLDA, TFA, and TPLDA, as well as the fusion of these classifiers on collections of video data. Experiments on the VidTIMIT+UMIST and the FERET datasets have shown that fusion of multiple classifiers improves face recognition across poses, given that the individual classifiers have similar performance. This proves that different probabilistic latent variable models learn statistical properties of the data that are complementary (not redundant). Furthermore, fusion across multiple images has also been shown to produce better perfomance than recogition using single still image.

Indoor Localization Using Visible Light Via Fusion of Multiple Classifiers

A multiple classifiers fusion localization technique using received signal strengths (RSSs) of visible light is proposed, in which the proposed system transmits different intensity modulated sinusoidal signals by LEDs and the signals received by a Photo Diode (PD) placed at various grid points. First, we obtain some {\emph{approximate}} received signal strengths (RSSs) fingerprints by capturing the peaks of power spectral density (PSD) of the received signals at each given grid point. Unlike the existing RSSs based algorithms, several representative machine learning approaches are adopted to train multiple classifiers based on these RSSs fingerprints. The multiple classifiers localization estimators outperform the classical RSS-based LED localization approaches in accuracy and robustness. To further improve the localization performance, two robust fusion localization algorithms, namely, grid independent least square (GI-LS) and grid dependent least square (GD-LS), are proposed to combine the outputs of these classifiers. We also use a singular value decomposition (SVD) based LS (LS-SVD) method to mitigate the numerical stability problem when the prediction matrix is singular. Experiments conducted on intensity modulated direct detection (IM/DD) systems have demonstrated the effectiveness of the proposed algorithms. The experimental results show that the probability of having mean square positioning error (MSPE) of less than 5cm achieved by GD-LS is improved by 93.03\% and 93.15\%, respectively, as compared to those by the RSS ratio (RSSR) and RSS matching methods with the FFT length of 2000.

SAR Target Recognition Based on Multi-feature Multiple Representation Classifier Fusion

SAR Target Recognition Based on Multi-feature Multiple Representation Classifier Fusion

Channel Division Based Multiple Classifiers Fusion for Emotion Recognition Using EEG signals

With the rapid development of computer technology, pervasive computing and wearable devices, EEG-based emotion recognition has gradually attracted much attention in affecting computing (AC) domain. In this paper, we propose an approach of emotion recognition using EEG signals based on the weighted fusion of multiple base classifiers. These base classifiers based on SVM are constructed using a channel division mechanism according to the neuropsychological theory that different brain areas are differ in processing intensity of emotional information. The outputs of channel base classifiers are integrated by a weighted fusion strategy which is based on the confidence estimation on each emotional label by each base classifier. The evaluation on the DEAP dataset shows that our proposed multiple classifiers fusion method outperforms individual channel base classifiers and the feature fusion method for EEG-based emotion recognition.

Predicting Non Performing Loan of Business Bank with Data Mining Techniques

The non-performing loans (NPL) prediction plays an important role in business bank. However, there is still a large gap between the requirement of prediction performance and current techniques. In this paper data mining approaches is used to predict the NPL. Both macroeconomic and bank-specific variables are collected to form the feature set firstly. Based on selected features, the study firstly applies single basic classifiers such as decision tree, k nearest neighbors and support vector machine (SVM) to model the problem of NPL. Bagging and AdaBoost are described in this paper as two different method of multiple classifier fusion, to build prediction models. In this experiment, non-performing loans data with 96 features and 10415 instances of a business bank is collected. F-mean and The Area under the ROC Curve (AUC) are considered as metrics of classification. The results illustrate that multiple classifier fusion algorithms outperform single basic classifier. The model built by multiple classifiers fusion can produce better prediction results. Furthermore, the AdaBoost method performs much better than bagging method in processing NPL.

Fault Diagnosis Model Based on Multi-level Information Fusion for CNC Machine Tools

The difficulty of CNC machine tools fault diagnosis is bigger than other general equipments because of the complex structure and the coupling among subsystems. The fault diagnosis model based on multi-level information fusion and hybrid intelligence is studied to improve reliability of fault diagnosis. Information from built-in sensors is used to monitor the status of CNC machine tools. The diagnosis principles of internal parameters-motor current, torque, temperature and following error are analyzed. Internal information and external sensors are two main sources which provide data to diagnosis. In order to detect effective fault signal, features of time domain, frequency domain and time-frequency domain are extracted. All these features constitute the feature set. The features are selected by the method of Kernel Principal Component Analysis (KCPA). Then the sensitive feature set is obtained. The method of multiple classifier fusion based on fuzzy comprehensive evaluation is researched. The determination method of weight based on information entropy is proposed. This diagnosis model has been tested feed system mechanical fault diagnosis of CNC machine tools and the results show which is effective and versatile.

Predicting non-performing loan of business bank by multiple classifier fusion algorithms

This paper uses multiple classifier fusion methods to predict the non-performing loans (NPL) of business bank. Both macroeconomic and bank-specific variables are collected to form the feature set firstly. Based on selected features, the study applies bagging and AdaBoost algorithms, which are described in this paper as two different method of multiple classifier fusion, to build prediction models. To contrast the classification performances, several basic strong classifiers such as decision tree, k nearest neighbors and support vector machine (SVM) are also adopted. In this experiment, non-performing loans data with 96 features and 10415 instances of a business bank is collected. F-mean and the Area under the ROC Curve (AUC) are considered as metrics of classification performances. The results illustrate that multiple classifier fusion algorithms outperform single basic classifier. Furthermore, the AdaBoost method performs much better than bagging method in processing NPL data of business bank.