Review of Classifier Combination Methods

Tulyakov, Sergey; Jaeger, Stefan; Govindaraju, Venu; Doermann, David

doi:10.1007/978-3-540-76280-5_14

Cited by 160 publications

(106 citation statements)

References 53 publications

(67 reference statements)

Supporting

Mentioning

102

Contrasting

Unclassified

Order By: Relevance

“…The goal of combining classifiers is to conceive a more efficient one which operates on the same data as the base classifiers and separates the same type of classes [21]. As stated in [21] if we denote the score assigned to the class i, by the classifier j : Sij , Therefore, a combination rule is a function f , and the final combination of the scores of the class i is in (3):…”

Section: Combination Of Classifiersmentioning

confidence: 99%

“…As stated in [21] if we denote the score assigned to the class i, by the classifier j : Sij , Therefore, a combination rule is a function f , and the final combination of the scores of the class i is in (3):…”

Section: Combination Of Classifiersmentioning

confidence: 99%

“…In this work we have two classifiers of the same type (neural networks) that are trained with different features. So we are interested in non-ensemble combinations [21].…”

Section: Combination Of Classifiersmentioning

confidence: 99%

See 2 more Smart Citations

Spatial and Textural Aspects for Arabic Handwritten Characters Recognition

Boulid¹,

Souhar²,

Ouagague³

2018

IJIMAI

View full text Add to dashboard Cite

The purpose of the present paper is the recognition of handwritten Arabic characters in their isolated form. The specificity of Arabic characters is taken into consideration, each of the proposed feature extraction method integrates one of the two aspects: spatial and textural. In the first step, a modified Bitmap Sampling method is proposed, which converts the character's images into a binary Matrix and then constructs a Mask for each class. A matching rate is used between the input binary matrix and the masks to determinate the corresponding class. In the second step we investigate the use of an Artificial Neural Network as classifier with the binary matrices as features and then the histograms of Local Binary Patterns to capture the texture aspect of the characters. Finally, the results of these two methods are combined to take into consideration both aspects at the same time. Tested on the Arabic set of the Isolated Farsi Handwritten Character Database, the proposed method has 2.82% error rate.

show abstract

Section: Combination Of Classifiersmentioning

confidence: 99%

Section: Combination Of Classifiersmentioning

confidence: 99%

See 1 more Smart Citation

Spatial and Textural Aspects for Arabic Handwritten Characters Recognition

Boulid¹,

Souhar²,

Ouagague³

2018

IJIMAI

View full text Add to dashboard Cite

show abstract

“…Theoretical and empirical evidence suggests that combining the responses of several accurate and diverse classifiers can enhance the overall accuracy and reliability of a pattern classification system [4,10]. Despite reducing information to binary decisions, combining responses at the decision level, in the Receiver Operating Characteristic (ROC) space, allows to combine across a variety of classifiers trained with different hyperparameters, feature subsets and initializations.…”

Section: Introductionmentioning

confidence: 99%

Incremental Boolean Combination of Classifiers

Khreich

Granger

Miri

et al. 2011

Multiple Classifier Systems

View full text Add to dashboard Cite

Abstract. The incremental Boolean combination (incrBC ) technique is a new learn-and-combine approach that is proposed to adapt ensemblebased pattern classification systems over time, in response to new data acquired during operations. When a new block of training data becomes available, this technique generates a diversified pool of base classifiers from the data by varying training hyperparameters and random initializations. The responses of these classifiers are then combined with those of previously-trained classifiers through Boolean combination in the ROC space. Through this process, an ensemble is selected from the pool, where Boolean fusion functions and thresholds are adapted for improved accuracy, while redundant base classifiers are pruned. Results of computer simulations conducted using Hidden Markov Models (HMMs) on synthetic and real-world host-based intrusion detection data indicate that incrBC can sustain a significantly higher level of accuracy than when the parameters of a single best HMM are re-estimated for each new block of data, using reference batch and incremental learning techniques. It also outperforms static fusion techniques such as majority voting for combining the responses of new and previously-generated pools of HMMs. Pruning prevents pool sizes from increasing indefinitely over time, without adversely affecting the overall ensemble performance.

show abstract

“…On the one hand, more and more applications require the integration of models and experts that come from different sources (human expert models, data mining or machine learning models, etc.). On the other hand, it has been shown that an appropriate combination of several models can give better results than any of the single models alone [26] [34], especially if the base classifiers are diverse [28].…”

Section: Introductionmentioning

confidence: 99%

Model Integration in Data Mining: From Local to Global Decisions

Sanjuán¹

View full text Add to dashboard Cite

A mis padres. A mis hermanas.A María. AcknowledgmentsEcho la vista atrás y me acuerdo de aquel primer día en que empecé mi carrera en la UPV. Recuerdo que pensé: "al menos durante los 5 próximos años tendré que hacer este mismo recorrido todos los días". Y así fue, pero lo que no pensaba es que después de la ingeniería vendría el doctorado, así que he seguido haciendo ese mismo camino, pero en coche en vez de en tranvía. Esta tesis, que tiene entre sus manos, en la pantalla de su ordenador o dispositivo móvil, es la culminación de todos estos años de investigación con un grupo de personas con las que muchas veces he compartido más tiempo que con mi familia y a los que les doy las gracias. Con ellos he compartido esos momentos de frustración en los que no sale nada y en los que lo más fácil sería abandonar, pero siempre me han apoyado para seguir adelante. Y, por supuesto, también hemos compartido los buenos momentos de alegría y satisfacción, que al final son los que más se recuerdan. Todo esto para decirles que si no fuese por ellos no estarían leyendo lo que están leyendo.En particular, gracias a Salvador Lucas que empezó toda esta aventura enviándome un correo para una beca. A María Alpuente, líder del grupo ELP, que me dio la oportunidad de pertenecer a este grupo. A mis directores de tesis: José, María José y César, por toda la ayuda que me han brindado y por su paciencia y dedicación en las revisiones de los trabajos que hemos realizado juntos. A Bea por todo el tiempo que hemos compartido juntos, por todo lo que hemos llorado y sobre todo por todo lo que hemos reído. A Vicent por todos sus buenos consejos. A Josep por su "introducción a la investigación". A Raquel por nuestras charlas. A Tama, Sonia, Rafa, Michele y Vesna porque las comidas de esaépoca llegaron a ser el momento más divertido del día. Y a todos los compañeros que han pasado por el grupo ELP durante estos años y que no enumero porque prefiero no olvidarme de ninguno.Y en especial, muchas gracias a mis padres, a mis hermanas y a mi novia María que saben todo lo que ha costado llegar hasta este momento y que a partir de hoy van a estar un poco más orgullosos de su hijo/hermano/novio.No se que me deparará el futuro, eso sólo el tiempo lo dirá, pero de lo que estoy seguro es que estaría encantado de seguir recorriendo cada día este mismo camino.Junio 2012, Antonio Bella. AbstractMachine Learning is a research area that provides algorithms and techniques that are capable of learning automatically from past experience. These techniques are essential in the area of Knowledge Discovery from Databases (KDD), whose central stage is typically referred to as Data Mining. The KDD process can be seen as the learning of a model from previous data (model generation) and the application of this model to new data (model deployment). Model deployment is very important, because people and, very especially, organisations make decisions depending on the results of the models. Usually, each model is learned independently from the others, trying to obtain the best...

show abstract

Review of Classifier Combination Methods

Cited by 160 publications

References 53 publications

Spatial and Textural Aspects for Arabic Handwritten Characters Recognition

Spatial and Textural Aspects for Arabic Handwritten Characters Recognition

Incremental Boolean Combination of Classifiers

Model Integration in Data Mining: From Local to Global Decisions

Contact Info

Product

Resources

About