An Analysis on the Purpose of Evolutionary Technique In Pattern Classification
A Comparative Study of Soft Computing Techniques in Pattern Classification
by Vivek Kumar*,
- Published in International Journal of Information Technology and Management, E-ISSN: 2249-4510
Volume 6, Issue No. 1, Feb 2014, Pages 0 - 0 (0)
Published by: Ignited Minds Journals
ABSTRACT
The pattern classification is a crucial errand of the vast majority ofthe human exercises where the choice or figure is made on the premise of rightnow accessible data and a classification technique is then some formal methodfor more than once making such judgments in new circumstances. There are aassortment of genuine issues, for example Data mining, Pattern Recognition,Image processing, Voice distinguishment and so on for which, singularclassification strategies are either insufficient or extremely dull to apply.Delicate computing strategies have been produced to fill this crevice and havepicked up expanding fame in the later years.
KEYWORD
evolutionary technique, pattern classification, human activities, data mining, pattern recognition, image processing, voice recognition, classification methods, soft computing, real problems
INTRODUCTION
Evolutionary computing reproduces the Darwinian standard of natural selection where it utilizes the standards of rivalry, legacy furthermore variety inside a population and the primary idea of natural selection is that the fittest people in a population survive and the weakest people die. Evolutionary algorithms are dependent upon heuristic look methods inside the extensively characterized space of artificial intelligence which utilizes the natural advancement methodology. Most human exercises could be characterized as a type of complex critical thinking. Therefore, critical thinking by a machine system is acknowledged to be an example of artificial intelligence, where the assignment comprises of supervising data and scanning for solutions. Given a representation of a issue and a depiction of a perfect solution, the objective of inquiry is to find a solution equivalent to or shut the perfect solution. Much of the time, choice making and critical thinking are formed as optimization issues. The part of the hunt method is to uncover the best solution around conceivable plan B, advancing for solution quality. Evolutionary algorithms generally utilize the accompanying things:
- A set of unique solutions (Candidate) called population.
- A fitness capacity that assesses and allots every singular a score on the other hand fitness quality.
- Transformation specialists that prepare posterity people from guardian people, executing the notion of legacy through stochastic variety.
- A stochastic selection method for selecting people with better fitness to prepare offspring.`
Evolutionary algorithms are broadly useful stochastic pursuit methods reenacting natural selection and development in the living planet. Evolutionary algorithms contrast from other optimization methods, for example Hill-Climbing and Simulated Annealing, in the way that Evolutionary algorithms uphold a population of potential (or hopeful) solutions to an issue, and not only one solution. Because of its population-based nature, Evolutionary Algorithms can abstain from being trapped in a neighborhood best and subsequently can regularly find global optimal solutions. Hence, Evolutionary Algorithms might be seen as global optimization algorithms. Nonetheless, it ought to be noted that Evolutionary Algorithms might neglect to merge to a global best. Evolutionary Algorithms have effectively been connected to a wide mixture of optimization issues, for instance: image processing, pattern recognition, booking, engineering design, etc.
CONCEPT OF GENETIC ALGORITHM
Initially pioneered by John Holland in the 60s, Ga, have been generally contemplated with investment, tested and connected in numerous fields in science and engineering planets. Ga is an evolutionary calculation, which enhances a fitness capacity to uncover the solution of an issue.
Available online at www.ignited.in Page 2
Distinctive evolutionary algorithms have been utilized for Fs. In a common Ga, each chromosome speaks to a prospective solution of the issue. The issue is connected with a fitness capacity — higher fitness alludes to a better solution. The set of chromosomes is known as a population. The population experiences a rehashed set of emphasess (or eras) with hybrid and transformation operations to find better solutions. At a certain fitness level or after a certain number of cycles, the technique is halted and the chromosome giving the best solution is protected as the solution of the issue. Gas are guided irregular look strategies used to search for parameters that give an exceptional solution to an issue. Basically they are nothing more than learned surmising. The "instruction" originates from knowing the suitability of past applicant solutions and the "surmising" originates from joining the fitter endeavors to advance a progressed solution.
Figure : Flow diagram of Optimization process of GA The persuasion for Gas hailed from nature and survival of the fittest. In a population, every distinct has a set of aspects that figure out how decently suited it is to nature. Survival of the fittest infers that the "fitter" people are less averse to survive and have a more amazing risk of passing their "exceptional" features to the following era. In sexual propagation, if the best features of every guardian are inherited by their posterity, another singular will be made that might as well have an enhanced likelihood of survival. This is the methodology of development. A key component in Ga is the selection of a fitness capacity that correctly quantifies the nature of hopeful solutions. A great fitness capacity empowers the chromosomes to successfully take care of a particular issue. Both the fitness capacity and solution representation are issue subordinate parameters. A poor selection of these two parameters will radically influence the execution of Ga. One issue identified with fitness capacities that may happen when Ga are utilized to advance combinatorial issues is the presence of focuses in the pursuit space that don't guide to plausible solutions. One solution to this issue is the expansion of a punishment capacity term to the definitive fitness capacity so chromosomes speaking to infeasible solutions will have a low fitness score, and as such, will vanish from the population.
FEATURE CHOICE USING GA
We know that all features that characterize a data point (class in most cases) for a classification problem may not be equally important; some features can be derogatory and may even have unfavorable influence on the task at hand. Feature selection techniques aim to discard the bad and irrelevant features from the available set of features. This reduction may improve the performance of classification, function approximation, and other pattern recognition systems in terms of speed, accuracy, and simplicity. We have applied the GA to optimize the subset of features as shown in Figure. Here initial population is a representation of different subsets of features with different lengths and combinations. GA has been applied for majority of optimization problems, including shortest path distance, traveling salesman problem, training of neural networks etc. Here, we apply GA to select a reduced subset of features. The objective function or the fitness function to be used in GA is the classification accuracy of the given classifier.
PARTICLE SWARM OPTIMIZATION
A Particle Swarm Optimizer (Pso) is a population-based stochastic optimization calculation. It is modeled on the social conduct of fowl herds. It could be effectively actualized and has been adequately connected to comprehend an extensive variety of optimization issues, for example ceaseless nonlinear and discrete optimization issues. In a Pso framework, a swarm of people (called particles)
Available online at www.ignited.in Page 3
fly through the seek space. Every molecule speaks to an applicant solution to the optimization issue. The position of a molecule is impacted by the best position went to without anyone else present (i.e. its own particular experience) and the position of the best molecule in its neighborhood (i.e. the knowledge of neighboring particles). The point when the neighborhood of a molecule is the whole swarm, the best position in the neighborhood is alluded to as the global best molecule, what's more the coming about calculation is alluded to as a gbest Pso. The point when littler neighborhoods are utilized, the calculation is by and large alluded to as a lbest Pso. The execution of every molecule (i.e. how close the molecule is from the global best) is measured utilizing a fitness capacity that shifts contingent upon the optimization issue. Pso is like Evolutionary Algorithms as in both methodologies are population-based and every distinctive has a fitness capacity. Besides, the changes of the people in Pso are moderately like the number-crunching hybrid admin utilized within Evolutionary Algorithms. Nonetheless, Pso is affected by the recreation of social conduct as opposed to the survival of the fittest. An alternate major distinction is that, in Pso, every distinctive profits from its history though no such instrument exists in Evolutionary Algorithms. The decision of neighborhood topology has a significant impact on the engendering of the best solution discovered by the swarm. Two regular neighborhood topologies are the star (or wheel) and ring (or loop) topologies. For the star topology one molecule is chosen as a hub, which is joined with all different particles in the swarm. Then again, all the different particles are just associated with the hub. For the ring topology, particles are masterminded in a ring. Every molecule has some number of particles to its correct and left as its neighborhood. As of late, proposed another Pso model utilizing a Von Neumann topology. For the Von Neumann topology, particles are associated utilizing a lattice arrange (2-dimensional grid) where every molecule is joined with its four neighbor particles.
ANT COLONY OPTIMIZATION SYSTEMS
Ant Colony Optimization Systems are another population-based stochastic approach which were first introduced by and Dorigo et al. to solve some difficult combinatorial optimization problems. ACO were inspired by the observation of real ant colonies. In real ant colonies, ants communicate with each other indirectly through depositing a chemical substance, called pheromone. Ants use, for example, pheromones to find the shortest path to food. This indirect way of communication via pheromones is called stigmergy. Using ACO, a finite size colony of artificial ants cooperates with each other via stigmergy to find quality solutions to optimization problems. Good solutions result from the cooperation of the artificial ants. ACO was applied to a wide range of optimization problems such as the traveling salesman problem, and routing and load balancing in packet switched networks with encouraging results. More details about Ant Systems and their applications can be found in Bonabeau et al. and Dorigo and Di Caro. ACO has been successfully applied to several combinatorial optimization problems to improve the performance.
CONCLUSION
This paper presents the Role of Evolutionary Approach in Pattern Classification, an introduction of evolutionary algorithms, An Overview of GA, How do the GA Works, and various GA’s Operators such as Selection, crossover and mutation, and Implementation of GA section covers Encoding, Solution Representation, Population size, Fitness Function. From the results presented in this paper, it can be concluded that the proposed schemes GA with nearest neighbor techniques and GA with PNN are the efficient techniques for pattern classification. These techniques have outperformed as compared to the conventional pattern classification techniques as well as other synergistic approaches published so far. This research proves that soft-computing techniques with proper synergistic approach can model the human brain better.
REFERENCES
- Derrig R. A. and Ostaszewski K. (1995). Fuzzy techniques of pattern recognition in risk and claim classification. J. Risk Insurance. 62, 447-482.
- Fukunaga K. (1989). Statistical Pattern Recognition. New York: Academic.
- Devroye L., Gyorfi L. and Lugosi G. (1996). A Probabilistic Theory of Pattern RecognitionSpringer-Verlag.
- Jain A., Duin R. and Mao J. (2000). Statistical Pattern Recognition : A Review. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(1), 4-37.
- Dorigo M. and Di Caro G. (1999). The Ant Colony Optimization Meta-Heuristic. New Methods in Optimization, D. Come, M. Dorigo and F. Glover, Eds., McGraw-Hill.
Janikow C., and Michalewicz Z. (1991). An Experimental Comparison of Binary and Floating Point
Available online at www.ignited.in Page 4
Representations in Genetic Algorithm. In Proceedings of the Fourth International Conference in Genetic Algorithms, Morgan Kaufmann. pp. 31-36.
- Liu, C.L. and Sako, H. (2006). Class-specific feature polynomial classifier forpattem classification and its application to handwritten numeral recognition, Pattern Recognition, vol. 39, pp. 669-681,2006.
- Rezaee M.R., Goedhart B., Lelieveldt B.P.F., Reiber J.H.C. (1999). Fuzzy feature selection, Pattern Recognition. 32,2011-2019.
- Krefiel U. and Schumann J. (1997). Pattern classification techniques based on function approximation”, in Handbook of Character Recognition and Document Image Analysis, H. Bunke and P.S.P. Wang (Editors), World Scientific, Singapore, pp. 49-78,1997.
- Mahamed G. H. Omran (2004). Particle Swarm Optimization Methods for Pattern Recognition and Image Processing, Phd Thesis, University of Pretoria.
- Pao Y.H. (1989). Adaptive Pattern Recognition and Neural Networks, Addison-Wesley, Reading, MA.
Misra B.B., Dehuri S., Dash P.K. Panda G. (2008a). A reduced and comprehensible polynomial neural network for classification, Pattern Recognition Letters of Elsevier Journal. 29 (2008), 1705-1712.