A review on instance ranking problems in statistical learning

Ranking problems, also known as preference learning problems, define a widely spread class of statistical learning problems with many applications, including fraud detection, document ranking, medicine, chemistry, credit risk screening, image ranking or media memorability. While there already exist reviews concentrating on specific types of ranking problems like label and object ranking problems, there does not yet seem to exist an overview concentrating on instance ranking problems that both includes developments in distinguishing between different types of instance ranking problems as well as careful discussions about their differences and the applicability of the existing ranking algorithms to them. In instance ranking, one explicitly takes the responses into account with the goal to infer a scoring function which directly maps feature vectors to real-valued ranking scores, in contrast to object ranking problems where the ranks are given as preference information with the goal to learn a permutation. In this article, we systematically review different types of instance ranking problems and the corresponding loss functions resp. goodness criteria. We discuss the difficulties when trying to optimize those criteria. As for a detailed and comprehensive overview of existing machine learning techniques to solve such ranking problems, we systematize existing techniques and recapitulate the corresponding optimization problems in a unified notation. We also discuss to which of the instance ranking problems the respective algorithms are tailored and identify their strengths and limitations. Computational aspects and open research problems are also considered.

A Novel Multi-Criteria Decision Sorting Approach based on Chebyshev's Theorem for Supplier Classification Problem

One of the interesting features of Multi-Criteria Decision Making/ Multiple Attribute Decision Making (MCDM/ MADM) is that a number of techniques that can be used to solve the same problem. In general, three common categories of decision problems are choice problem, ranking problem, and sorting problem. While, the issue of choice and ranking problems is more emphasized in MCDM/ MADM, but the literature weakly consider sorting problems. Several solutions for the above problem are suggested (i.e., Flow sort, AHP-Sort, ELECTRE Tri, etc.). Theoretically, there is no reason to be limited to these techniques. Hence, in this paper we propose a novel multi-criteria sorting method that is based on Chebyshev’s theorem. More specifically, different from other studies on MCDM sorting problems, which put more emphasis on the extension of new models, this work attempts to present a general framework using the Chebyshev’s inequality, to transform the results of conventional MCDM models from ranking format to sort mode. Finally, the proposed approach is compared with three existed models. Compared results show that the proposed method is efficient and the results are stable.

A human-intuitive fuzzy ranking approach with spread and skewness factors

It would be hard to deny the importance of fuzzy number ranking in fuzzy-based applications. The definition of fuzzy ranking, however, evades an easy description due to the overlapping of fuzzy sets. While many researchers have addressed this subject, close examination reveals that their results suffer from one or more shortcomings such as image-ranking problems or ranking two equally embedded fuzzy numbers with the same centroid and different spreads. This paper proposes a new fast and straightforward computational approach to ranking fuzzy numbers that aims to overcome such problems. The proposed approach considers several important factors such as spread, skewness and center, in addition to human intuition. Further, the proposed ranking approach involves a composition of these factors as demonstrated in the several examples provided and in comparison with other existing approaches.

Revisiting Feature Ranking Methods using Information-Centric and Evolutionary Approaches: Survey

: Feature ranking can have a severe impact on the feature selection problem. Feature ranking methods refer to the structure of features that can accept the designed data and have a positive effect on the quality of features. Moreover, accessing useful features helps in reducing cost and improving performance of a feature ranking algorithm. There are numerous methods for ranking the features are available in literature. The developments of the past 20 years in the domain of knowledge research has been explored and presented in terms of relevance and various known concepts of feature ranking problems. The latest developments are mostly based on the evolutionary approaches which broadly include variations in ranking, mutual information, entropy, mutation, parent selection, genetic algorithm etc. For a variety of algorithms based on differential evolution, it observed that the suitability of the mutation operator is extremely important for feature selection but other operators can be considered. Therefore, the special emphasis of various algorithms is observing and reviewing the algorithms and to find new research directions.: The general approach is to do a rigorous collection of articles first and then obtain the most accurate and relevant data followed by the narrow down of research questions. Research is based on the research questions. These are reviewed in four phases : designing the review, conducting the review, analysis, and then writing the review. Threats to Validity is also considered with research questions. In this paper, many feature ranking methods have been discussed to find further direction in feature ranking and differential evolution. A literature survey is performed on 93 papers to find out the performance in relevance, redundancy, correlation with differential evolution. Discussion is suitable for cascading the direction of differential evolution in integration with information-theoretic, entropy and sparse learning. As differential evolution is multi-objective in nature so it can be incorporated with feature ranking problems. The survey is being conducted on many renowned journals and is verified with their research questions.Conclusions of the survey prove to be essential role models for multiple directions of a research entity. In this paper, a comprehensive view on the current-day understanding of the underlying mechanisms describing the impact of algorithms and review current and future research directions for use of evolutionary computations, mutual information and entropy in the field of feature ranking is complemented by the list of promising research directions. However, there are no strict rules for the pros and cons of alternative algorithms.

Aggregating the results of benevolent and aggressive models by the CRITIC method for ranking of decision-making units: A case study on seven biomass fuel briquettes generated from agricultural waste

The ranking of decision-making units (DMUs) is one of the main issues in data envelopment analysis (DEA). Hence, many different ranking models have been proposed. However, each of these ranking models may produce different ranking results for similar problems. Therefore, it is wise to try different ranking models and aggregate the results of each ranking model that provides more reliable results in solving the ranking problems. In this paper, a novel ranking method (Aggregating the results of aggressive and benevolent models) based on the CRITIC method is proposed. To prove the applicability of the proposed ranking method, it is examined in three numerical examples, six nursing homes, fourteen international passenger airlines and seven biomass materials for processing into fuel briquettes. First, benevolent and aggressive models were used to calculate the efficiency rating for each DMU. As a result, the decision matrix was generated. In the decision matrix, the results of benevolent and aggressive models were viewed as criteria and DMUs were viewed as alternatives. Then, the weights of each criterion were generated by the CRITIC method. Finally, each DMU was ranked. In a comparative analysis, the proposed method can lead to achieving a more reliable decision than the method which is based on a stand-alone method.

Perceptron Ranking Using Interval Labels with Ramp Loss for Online Ordinal Regression

Due to its wide applications and learning efficiency, online ordinal regression using perceptron algorithms with interval labels (PRIL) has been increasingly applied to solve ordinal ranking problems. However, it is still a challenge for the PRIL method to handle noise labels, in which case the ranking results may change dramatically. To tackle this problem, in this paper, we propose noise-resilient online learning algorithms using ramp loss function, called PRIL-RAMP, and its nonlinear variant K-PRIL-RAMP, to improve the performance of PRIL method for noisy data streams. The proposed algorithms iteratively optimize the decision function under the framework of online gradient descent (OGD), and we justify the algorithms by showing the order preservation of thresholds. It is validated in the experiments that both approaches are more robust and efficient to noise labels than state-of-the-art online ordinal regression algorithms on real-world datasets.

A kernel machine for hidden object-ranking problems (HORPs)

Hidden Object-Ranking Problems (HORPs) are object-ranking problems stated as classification or instance-ranking problems. There exists so far no dedicated algorithm for solving them properly and HORPs are usually solved as if they were classification (multi-class or ordinal) or instance-ranking problems. In the former case, item-related ordinal information is negated and only class-related information is retained; in the latter case, item-related ordinal information is considered, but in a way that emphasizes class-related information, so that the items are not only sorted but also clustered. We propose a kernel machine that allows retaining item-related ordinal information while avoiding emphasizing class-related information. We show how this kernel machine can be implemented with standard optimization libraries provided slight modifications on the original kernel. The proposed approach is tested on Tahitian pearls quality assessment and compared with four other classical methods. It yields better results (93.6% ± 3.9% of correct predictions without feature selection, 94.3% ± 3.4% with feature selection) than the best of the other tested methods (91.3% ± 3.4% and 92.6% ± 4.3% without and with feature selection for the instance-ranking approach), this improvement being significant (p-value < 0.05). Moreover, this method exhibits no significant difference in the results with and without feature selection (p-value = 0.33), which may be a hint that its learning bias fits the problem well and can thus alleviate the data preprocessing workload.

HPOLabeler: improving prediction of human protein–phenotype associations by learning to rank

Motivation Annotating human proteins by abnormal phenotypes has become an important topic. Human Phenotype Ontology (HPO) is a standardized vocabulary of phenotypic abnormalities encountered in human diseases. As of November 2019, only &lt;4000 proteins have been annotated with HPO. Thus, a computational approach for accurately predicting protein–HPO associations would be important, whereas no methods have outperformed a simple Naive approach in the second Critical Assessment of Functional Annotation, 2013–2014 (CAFA2). Results We present HPOLabeler, which is able to use a wide variety of evidence, such as protein–protein interaction (PPI) networks, Gene Ontology, InterPro, trigram frequency and HPO term frequency, in the framework of learning to rank (LTR). LTR has been proved to be powerful for solving large-scale, multi-label ranking problems in bioinformatics. Given an input protein, LTR outputs the ranked list of HPO terms from a series of input scores given to the candidate HPO terms by component learning models (logistic regression, nearest neighbor and a Naive method), which are trained from given multiple evidence. We empirically evaluate HPOLabeler extensively through mainly two experiments of cross validation and temporal validation, for which HPOLabeler significantly outperformed all component models and competing methods including the current state-of-the-art method. We further found that (i) PPI is most informative for prediction among diverse data sources and (ii) low prediction performance of temporal validation might be caused by incomplete annotation of new proteins. Availability and implementation http://issubmission.sjtu.edu.cn/hpolabeler/. Contact [email protected] Supplementary information Supplementary data are available at Bioinformatics online.