Selecting an Optimal Compound of a University Research Team by Using Genetic Algorithms

Size: px
Start display at page:

Download "Selecting an Optimal Compound of a University Research Team by Using Genetic Algorithms"

Transcription

1 Selecting an Optimal Compound of a University Research Team by Using Genetic Algorithms Florentina Alina Chircu 1 (1) Department of Informatics, Petroleum Gas University of Ploiesti, Romania chircu_florentina@yahoo.com Abstract The latest economic situation determines an increased attention to efficient and rational use of productive resources of capital and labour. In this context, higher education institutions are trying to encourage building convenient research teams, taking into account the fact that research is dependent upon the individuals. The members of a research team must be chosen considering the importance of their knowledge for the proposed project but also the significance of the project for the individuals career. In this paper it is presented an application that proposes the implementation of genetic algorithms in this area. The application aims to identify the best compound of a research team by choosing the most suitable individuals from different university departments in order to increase the productivity and to minimize the cost concerning time and resources. Keywords: Research team, Genetic algorithms, Artificial intelligence Introduction In recent times the development of the research discipline is obviously growing. The demands of the research organisation and the results expected are more and more complex. The latest economic situation determines an increased attention to efficient use of productive resources of capital and labour. To accomplish the current requirements, a method to optimally use all the resources (financial, human and physical) needs to be identified [4]. In this context higher education institution are making significant efforts to encourage the development of the research area. An essential fact that they take into account is represented by the importance of building the most convenient and efficient research teams in a given situation. The results obtained by a research team depend first on individuals and second on other factors. It was noted that the projects with the best performances in the research activity are composed by carefully chosen individuals, with high career motivation that are perfect for the selected job [4]. In this paper is presented the implementation of a genetic algorithm which aims to identify the best compound of a research team by choosing the most suitable individuals from different university departments in order to increase the productivity and to minimize the cost concerning time and resources. The results returned by the application are represented by a proposal for the team compound by choosing individuals which are the most appropriate for the job, considering a set of restrictions. Genetic Algorithms Genetic Algorithms represent an area of the artificial intelligence that has known a great development. They represent an evolutionary search technique used with the intention to identify

2 The 5 th International Conference on Virtual Learning ICVL an approximate solution for optimization and search problems. Genetic Algorithms are inspired from the evolutionary biology and they bring up techniques that simulate the natural population evolution, such as inheritance, mutation, selection and crossover (also called recombination) [2]. Genetic Algorithms represent a computer simulation of the natural evolution. For a population of individuals who represent abstract representation of candidate solutions to an optimization problem is simulated an evolution following several steps in order to obtain higher quality solutions [1]. The evolution toward better solution is guaranteed by the fact that only the strongest individuals will be able to adapt to the virtual environment and respect all the imposed restrictions, and so they will survive the evolution. The components of a genetic algorithm are [2]: A representation of optimisation problem solutions as chromosomes; A method to generate initial populations of potential solutions; A method to evaluate each individual performance (fitness function); Runtime parameters (population size, crossover possibility, mutation probability and evolution interval). The algorithm is defined by the following steps [3]: 1. Complete or partially randomly generation of initial population of candidate solutions; 2. If the population size is higher than the default parameter, return the best solution. 3. Calculate the fitness function for each individual; 4. Apply the genetic operators and generate the new population: Select the individuals with the highest fitness function to become parents; Achieve new individuals by parents combination; Apply the mutation operation if necessary; 5. Repeat from Step 2. The fitness function represents a measure of the solution quality. This function calculates the individuals performance based on several criteria. Based on this fitness function, the best individuals are selected from the current population and recombined to obtain the new individuals which will be inserted into the new population. The crossover operation proposes a way of obtaining the propagation of the best genetic material in order to increase the quality of the candidate solution populations. This genetic operator presumes the recombination of two chromosomes with the purpose of obtaining new individuals to be included in the new population. The mutation is represented by a small chromosome alteration applied to one or more genes. The genetic algorithm reaches the end when the maximum number of generation members has been reached. The individuals with the highest fitness function are selected from the current population and are returned as solutions for the optimisation problem. Genetic Algorithms may find their application in many fields as bioinformatics, chemistry, mathematics, physics, engineering, computational science and others. They frequently have application in problems as [2]: Classification problems; Tasks planning; Network flow problems; Real time optimization; Prediction systems (economical, geological, structural and others); Neural network design; Robots trajectories determination.

3 382 University of Bucharest and University of Medicine and Pharmacy Târgu-Mureş Description of the Proposed Genetic Algorithm The application described in this paper aims to help increase the university research management by presenting a method to optimally use all the human resources available. This application intends to identify the best compound of a research team by choosing the most suitable individuals from different university departments in order to increase the productivity and to minimize the cost concerning time and resources. There are considered to be available 3 departments, each department consisting in different number of members. Each member has associated a coefficient that indicates the importance that the proposed project has regarding the individuals career, but also the significance that the individuals knowledge has on the subject of the proposed project. All the necessary data concerning to the number of the individuals on each department, to the individuals identification number and to the coefficient that measures the importance of the individual for the proposed project are presented in a text file named date_in.txt who is reproduced in Table 1. The matching coefficient is a number between 0 and 100. Genetic codification of candidate solution is: (S 1 S 2 S 3 S 4 S 5 S 6 S 7 S 8 S 9 S 10 ) where S i represents a structure that codifies the individuals identification number, the department that he s a part of and the coefficient that identifies the importance of the individual for the proposed project. The implementation of this genetic codification in C++ Builder is: typedef struct departament { int dep; //the department identification number int id_ind; // the individual identification number int coef; // the individual coeficient int free; //the individuals availability } d; typedef struct cromozom { int d[12];//genetic codification of candidate solution int fitness; //fitness function value } ind; Each individual s performance is measured by the value of the fitness function. This functions is initialised with 100 and it is modified according to a mathematical formula considering the importance of the individual for the proposed project and the fact that he is available to participate, to be a part of the team. Genetic operators used in this application are: elitist selection, two-point crossover, translation mutation. The elitist selection consist of choosing future parents from the current population, the selection criterion is the individuals performance. The current population individuals are sorted according to the fitness function value and only the best individuals are chosen to become parents. The two-point crossover is a genetic operator that is used to combine the genetic material of two individuals called parents with the purpose of obtaining new individuals, which will be included in the new population. The translation mutation is a genetic operator that consists in performing a small alteration in the individuals structure. A set of genes with random length is selected and is moved back or forward with a random number of positions.

4 The 5 th International Conference on Virtual Learning ICVL The parameters corresponding to the genetic algorithm which can be set using the application interface are: Initial population size (with values between 10 and 100); Maximum population size; Crossover probability (with values between 0.1 and 1.0); Mutation probability (with values between 0.01 and 0.2). The application will return using the user interface a solution for the proposed problem representing the individual with the highest fitness function. This solution represents a suggestion for the research team compound, considering all the given restrictions. The detailed results will be store in a data output text file named out.txt Experimental Results In order to test the application, it is presented a set of experimental data, which will simulate a real situation. There are considered to be available 3 departments, each department consisting in different number of members. The composition of each department, including the identification number of each member and individuals matching coefficient are presented in Table 1. Department Members number Table 1. Department compound Department members identification number Individuals coefficient In Table 2 are presented 3 sets of input parameters for the genetic algorithm. Considering the fact that the algorithm is based on a random generation of initial individuals solutions, there are repeated 30 tests for each set of parameters and identified the returned solutions.

5 384 University of Bucharest and University of Medicine and Pharmacy Târgu-Mureş Table 2. Experimental Values Set Values set Parameters Number of tests Initial population size Maximum population size Crossover probability Mutation probability The final results are synthesized in Table 3. The best performance average is calculated considering the highest fitness function values obtained during the tests. The worst performance average is calculate taking into account the lowest fitness function values obtained at the last generation all through the tests. The solution with the best performance obtained after the 90 tests is represented by the sequence :( [3,306, 90] [1, 102, 45] [3, 302, 65] [2, 208, 80] [3, 308, 70] [1, 104, 80] [2, 201, 75] [3, 304, 80] [2, 206, 60] [2, 203, 90]). When the solution is decode, it is represents a proposal for the best team compound, considering the individuals who are the most appropriate for the job and taking into account a set of restrictions: Department 3, member 306, coefficient value 90; Department 1, member 102, coefficient value 45, and so on. Table 3. Final Results Final results Best performances average at last generation Worst performances average at last generation Best performance 735 Number of solutions with the best performance 1 Best performance solution ([3,306, 90] [1, 102, 45] [3, 302, 65] [2, 208, 80] [3, 308, 70] [1, 104,80] [2, 201,75] [3, 304, 80] [2, 206, 60] [2, 203, 90]) Conclusions The latest economic situation imposes an increased attention to efficient use of productive resources of capital and labour even if we speak about the research area. The higher education institution are interested in the development of the research discipline and the competition for obtaining important financial resources for research require new quality demands regarding the results. Since the results depends upon the individuals, it is clearly known the importance of building the most convenient and efficient research teams in a given situation, by choosing the best individuals for the selected job, with high career motivation and corresponding knowledge in the proposed project theme. This paper proposes the implementation of a genetic algorithm in this area which intends to identify the best compound of a research team by choosing the most suitable individuals from different university departments in order to increase the productivity and to minimize the cost concerning time and resources.

6 The 5 th International Conference on Virtual Learning ICVL After 90 tests, it was identified the solution with the highest fitness function. The solution represents a proposal for the team compound by choosing individuals who are the most appropriate for the job, considering a set of restrictions. References Russel, S., Norving, P., Artificial intelligence: A Modern Approach, Prentice Hall, Oprea, M., Nicoara, S., Artificial intelligence, Petroleum Gas University of Ploiesti, Chircu, F., Using Genetic Algorithms to Increase the Quality of University Research Management, Proceedings of ICVL ***, University Research Management, accessed on