A Review of Pattern Recognition Using Statistics

Abstract

Pattern recognition\'s primary objective is supervised or classification techniques. The statistical method has been the most intensively researched and implemented in practice among the various frameworks in which classification technique has typically been developed. In recent times, neural network techniques and methods based on statistical learning theory have gotten a lot of attention. The following issues must be carefully considered when designing a recognition system: pattern classes, sensing environment, pattern representation, feature extraction and selection, cluster analysis, classifier design, learning, training, test sample selection, and performance evaluation. The overall challenge of detecting intricate forms with modifiable direction, position, and size has not been addressed despite just Fifty years of intensive research. Data mining, web browsing, recovery of multimedia content, face recognition, and cursive handwriting recognition are just a couple of minor new and emerging applications that require strong and fast pattern recognition methods. The intention of this review paper is to summarize and analyze some of the adopted approaches in various phases of a recognition system, as well as to identify areas of research and applications that are at the forefront of innovation in this exciting and challenging field.

Introduction

I. INTRODUCTION

The preponderance of children can recognize figures and characters by the age of five. Small, huge, scribbled, machine printed, or flipped characters are all easily identified by children. The characters might be scrawled on a confused foreground, folded paper, or perhaps even concealed in part. We take this capacity for granted until we are faced with the task of programming a system to do the same.

Pattern classification detection is the examination of where and how technologies can keep track of their surroundings, learn to recognize relevant patterns, and make well-informed judgments regarding the patterns' classifications. Despite over 50 years of research, establishing a standard machine pattern recognizer is still a distant possibility. Humans are the finest pattern recognizers in the majority of cases, but we also have no clue how humans achieve it. The research of Nobel Prize winner Herbert Simon, whose major finding was that pattern identification is crucial in most human decision-making activities, is highlighted by Ross [8]. The much more applicable patterns you possess, the smarter choices you'll make. This is exciting news that artificial intelligence proponents because computers can certainly be instructed to find patterns. Nevertheless, effective software applications that assist banks in consumer credit, doctors in illness diagnostics, and pilots in landing airplanes generally rely on pattern recognition in some way.

We need to concentrate a greater emphasis on pattern recognition in the curriculum. Our goal is to introduce pattern recognition as the main tactic to generate autonomous judgments using the information to contact, computers, and domain knowledge. The much more relevant patterns you possess, the better decisions you'll make. This is exciting news for artificial intelligence proponents because computers can certainly be taught to find patterns.

Pattern detection, characterization, classification, and clustering by systems are key challenges in a range of technical and scientific domains, encompassing biology, psychology, medicine, marketing, computer vision, artificial intelligence, and remote sensing, to mention very few.

II. LITERATURE REVIEW

They introduced a CCN approach in 2016, which was the first time it was used to extract drug-drug interactions [1]. To gather semantic understanding, positional and word embeddings are used; this CNN extraction strategy for drugs achieves an F-score of 69.75 percent.

The paper developed in 2017 that the best characteristics be used to determine the drug's combination. The suggested approach yields positive in terms of expectations, and it also took second place in the DDI extraction challenging task. The authors offered initial diagnostic care for a particular medication in 2019, which is extremely helpful to pharmacists [2]. In addition, in 2020, the authors include a recommendation on prescription medications to facilitate DDI analysis [3].

A hybrid approach depending on a Kalman filter, a convolutional neural network, and wavelet transformations were suggested by the researcher in [4]. Clustering methods were used in the hybrid model with relatively brief capacity and renewable energy forecasting.Hybrid models incorporating clustering-based wavelet processing and artificial neural network ( ann )outperformed conventional approaches and hybrid power model combinations, according to the findings. However, rather than genuine energy use patterns, this study's grouping was linked to particular zones.

The authors developed a DDI Exporter in 2006 which was the first online technique for detecting DDIs [5]. Users could use this tool to scan the Medline 2010 baseline database for DDIs employing keywords. Drugs and combinations are identified by the software.

Based on [6], the authors presented systematic methods for resolving analogical reasoning in drug-drug interaction articles in 2009. The resolution of anaphoras is used to record the effectiveness of extraction activities that use a rule-based linguistic approach. To commence, the MMTx (MetaMap Transfer) text is processed employing UMLS models to accomplish tokenization, sentence splitting, linking, POS-tagging, and character segmentation. The basic components are then passed on to the feature extraction phase. The corpus is intended for anaphora resolution for research purposes. Using DrugNerAR, the authors have tried to get a 73 percent performance for the baseline. DDI identification is a critical difficulty in medication planning and implementation.

The authors of [7] proposed a protracted energy forecasting system that predicts year-ahead demands for power systems using the Concentration Spatial Clustering of Implementations with Noise (DBSCAN) spatial clustering algorithm. The concentration clustering method utilizes the advantage of its inherent capacity to efficiently cope with data noise by removing outliers. Looking at historical yearly energy demand profiles, land use categories, and geographic information, DBSCAN was used to group comparable sub-zones. The predicted load values are finally produced using Non-Linear Auto-Regressive (NAR) neural network models. When compared to existing models like moving averages, grey theories, and regression analysis, they discovered that their proposed performance is better (LR). This approach, however, doesn't really deal with short-term load prediction.

III. PATTERN MATCHING

The three major aspects are central to the development of a recognition system: Data capture and preprocessing, statistical modeling, and decision-making are the three steps. The device(s), preprocessing methodology, representation strategy, and decision-making framework all rely on the area or subject. A well-defined and sufficiently restricted recognition problem (minimal intraclass variations and substantial interclass variations) is broadly acknowledged as yielding to a compact pattern representation and a clear and simple decision-making methodology.

Most pattern recognition programs incorporate and desire the capacity to study from a number of examples (training set). 1) templates comparison, 2) statistically classifying, 3) syntactic or structural corresponding, and 4) neural networks are the four most well-known techniques of pattern recognition.

Various interpretations with the same pattern identification systems may arise since these models are not truly independent. Attempts to construct hybrid systems involving various models have been made [9]. The accompanying is a comprehensive discussion and variety of various methodologies, which is summarized in Table 1.

TABLE 1. PATTERN MATCHING MODELS

IV. PATTERN MATCHING APPROACH

A. Template Matching

Proposed method is among the fundamental and oldest methods of pattern recognition. Matching is a pattern recognition procedure that determines the resemblance between two items of the same type (points, curves, or forms). A template (usually a 2D object) or a model of the patterns to be acknowledged is provided in template matching.

The detected pattern is contrasted to the stored template, taking into account all conceivable poses (translation and rotation) and scale adjustments. Template matching is technically demanding, although this approach is now more feasible with the emergence of better processors.

While rigorous template matching is effective in certain application domains, it brings with it a number of downsides. For example, if the patterns become distorted as a consequence of the imaging procedure, a change of perspective, or significant intraclass variances among the patterns, it would fail.

B. Statistical Approach

In the statistical model, each pattern is modeled as follows in a d-dimensional space using d variables or measurements. The objective is to select options that enable pattern vectors across separate classes to occupy compact and distinct parts of a d-dimensional subspace. How well patterns from different classes may be segregated influences the effectiveness of the representation space (feature set).The judgment boundaries in the statistical decision-theoretic approach are established by the distributions of each class's patterns, which must either be described or understood

C. Syntactic Approach

In many identification issues involving sophisticated forms, it is more logical to take a structured methodology, in which a pattern is seen to be made up of fundamental sub-patterns that are made up of an even simpler set of test cases. A formal analogy is developed between both the structure of patterns and the syntactic of a language in syntactic pattern recognition. The patterns are considered as sentences in a language, the primitives are considered as the language's alphabet, and the phrases are formed using grammar. As a consequence, a vast number of complicated patterns might well be expressed using only a few primitives and grammatical rules. The morphology of each pattern class must be determined using the training examples listed.

D. Neural Networks

Multithreaded computing systems, also including neural networks, are made up of a huge number of basic units with many interconnections. Learning, generalization, adaptivity, fault tolerance, distributed depiction, and computation are all efforts to use important components  in a network of calibrated directed graphs, in which the nodes are artificial nerve cells and the controlling behaviors (with weights) are interconnections among both hidden neurons and nerve cell inputs. Neural networks' main characteristics include their ability to learn complex nonlinear input-output connections, use sequential procedures, and adapt to inputs. For extracting features (using hidden layers) as well as classification, neural networks offer a new set of nonlinear techniques (e.g., multilayer perceptrons).Techniques for extracting initial feature features can also be applied to neural network topologies to improve (hardware) implementation efficiency. The bulk of well-known CNN designs is implicitly comparable to or similar to classic statistical pattern recognition algorithms, despite their obvious differences in basic principles. This link between neural networks and quantitative pattern recognition is also discussed by Ripley and Anderson et al.

Conclusion

It\'s not been easy to figure out which method is ideal for classifying patterns with the least amount of computing work, the shortest amount of time, and the best outcomes. Various types of pattern matching algorithms are explored in this review study. The subject of pattern recognition includes a wide variety of applications, including categorization, grouping, regression, sequence labeling, and parsing, among which this paper examines the methods of the most often used pattern recognition field, classification, and clustering. Pattern recognition using the classification strategy employs a labeled training set to classify unlabeled test data into the target category. The clustering methods, on the other hand, do not have a labeled set. To place the test set into the right cluster, they utilize a different metric, such as Euclidean distance.

References

[1] S. Liu, B. Tang, Q. Chen, and X. Wang, \"Drug-Drug Interaction Extraction via Convolutional Neural Networks,\" Comput. Math. Methods Med., vol. 2016, pp. 1–8, Jan. 2016, doi: 10.1155/2016/6918381. [2] M. Carpenter, H. Berry, and A. L. Pelletier, \"Clinically Relevant Drug-Drug Interactions in Primary Care.,\" Am. Fam. Physician, vol. 99, no. 9, pp. 558–564, May 2019. [3] S. Sudsakorn, P. Bahadduri, J. Fretland, and C. Lu, \"2020 FDA Drug-drug Interaction Guidance: A Comparison Analysis and Action Plan by Pharmaceutical Industrial Scientists.,\" Curr. Drug Metab., vol. 21, no. 6, pp. 403–426,2020,doi:10.2174/1389200221666200620210522. [4] H. H. H. Aly, ‘‘A proposed intelligent short-term load forecasting hybrid models of ANN, WNN and KF based on clustering techniques for smart grid,’’ Electr. Power Syst. Res., vol. 182, May 2020, Art. no. 106191, doi: 10.1016/j.epsr.2019.106191. [5] I. Sa?nchez-Serrano, \"Success in translational research: lessons from the development of bortezomib.,\" Nature Reviews. Drug discovery, vol. 5, no. 2. England, pp. 107–114, Feb. 2006, doi: 10.1038/nrd1959. [6] M. Malvaez, R. M. Barrett, M. A. Wood, and C. Sanchis-Segura, \"Epigenetic mechanisms underlying extinction of memory and drug-seeking behavior,\" Mamm. Genome, vol. 20, no. 9, pp. 612–623, 2009, doi: 10.1007/s00335-009-9224-3. [7] Z.Han, M.Cheng, F.Chen, Y.Wang,andZ.Deng, ‘‘Aspatialloadfore- casting method based on DBSCAN clustering and NAR neural network,’’ in Proc. J. Phys., Conf., 2020, vol. 1449, no. 1, Art. no. 012032, doi: 10.1088/1742-6596/1449/1/012032. [8] P.E. Ross, aFlash of Genius,o Forbes, pp. 98-104, Nov. 1998. [9] K.S. Fu, aA Step Towards Unification of Syntactic and Statistical Pattern Recognition,o IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 5, no. 2, pp. 200-205, Mar. 1983. [10] L. Devroye, L. Gyorfi, and G. Lugosi, A Probabilistic Theory of Pattern Recognition. Berlin: Springer-Verlag, 1996 [11] R.O. Duda and P.E. Hart, Pattern Classification and Scene Analysis, New York: John Wiley & Sons, 1973. [12] K.S. Fu, Syntactic Pattern Recognition and Applications. Englewood Cliffs, N.J.: Prentice-Hall, 1982 [13] G. Nagy, Candide\'s Practical Principles of Experimental Pattern Recognition,o IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 5, no. 2, pp. 199-200, 1983 [14] J. Anderson, A. Pellionisz, and E. Rosenfeld, Neurocomputing 2: Directions for Research. Cambridge Mass.: MIT Press, 1990.

Copyright

Copyright © 2022 Manvi Bohra, Abhishek Kumar Sarkar, Atul Jadli, Yuvraj Joshi, Sidharth Gupta. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.