Basketbolda oyuncu seçimi için sinirsel-bulanik karar destek sistemi
Makaleler > Tam Makale · 2009
Yazarlar: Serkan Ballı, Bahadır Karasulu, Aybars UĞUR, Mustafa Serdar KORUKOĞLU
İTÜ Dergisi D : Mühendislik, Uluslararası, 2009, cilt 8, ss 15-25
Özet
Decision Support Systems (DSS) are a kind of information systems and support effective decision making for multi-criteria problems. Artificial Neural Networks (ANN) are capable of general pattern classifications and recently used in decision support systems for assessment of pattern between criteria and alternatives. An ANN consists of a number of very simple and highly interconnected neurons which are used for modeling decision problems. In statistical decision making methods, decision making is the evaluation of alternatives according to some certain criteria and then preference scores are computed for each alternative. According to these numerical results, an alternative best meeting all the criteria is chosen and decision making is performed. Unlike these criteria described quantitatively, there are some problems involving qualitative criteria including some ambiguities and only expressed as linguistic so that they can not be precisely described. Multi-criteria decision problems may contain non-linear and uncertain criteria. In that case, fuzzy approaches are used for modeling non-linear attributes and informations. Uncertain criteria can be explained with fuzzy sets and membership functions. In daily life, there are a lot of complex problems which has vague and ambiguous information such as selection appropriate players for basketball game. For this example, information may contain uncertain terms like Player shots very well. In basketball game, there are a lot of criteria for selecting appropriate players and they can be divided into two groups: physical measurement criteria (vertical jump, height, weight etc.) and observation criteria (match observation, dripling, collective drills etc.). In this study, the aim is to develop a model which allows choosing the most appropriate basketball players. To realize this aim, it is studied how to evaluate together the numerical measurements of physical criteria and criteria that is analyzed via observation which is digitally immeasurable. For solving this problem, a concurrent Neuro-Fuzzy Decision Support System (NFDSS) model is developed. The model has been applied for twelve players in age between seven and fourteen by obtaining player's physical appropriateness measurement and observation criteria values. For all players, ten physical appropriateness criteria which are vertical jump, muscular power, action rapidity, reaction time (as to sound), height, weight, body mass index, body fat rate, endurance and anaerobic power have been measured in laboratory. Then values of observation criteria which are dripling, pass, shot, collective drills and observing match have been expressed by using linguistic values. For example, Player shots very well. In this example, the linguistic variable shot indicates a fuzzy set. Then all criteria values have been inserted to NFDSS for evaluation. NFDSS is built by using Matlab M-files. The next purpose is to investigate the appropriateness of the developed model for the system. NFDSS contains five layers. First layer is input layer and transmits external crisp physical measurement values to the next layer. Layer 2 is the fuzzification layer. This layer receives a crisp input and determines the degree to which this input belongs to fuzzy set. Then they are sent to ANN component (Layer 3) that comprises two ANN: ANN1 and ANN2. ANN1 is used for physical measurement values and ANN2 is used for observation values. During this process, physical appropriateness criteria values and observation criteria values have been combined by using criteria weights It has been ensured that using both significant criteria and the weights of criteria make results more sensitive. The Outputs of ANN1 and ANN2 are evaluated by Layer 4 which is called Fuzzy Inference System (FIS). In FIS, ANN1 and ANN2 values are joined by using rule-base. Rule-base has nine different rules. Some examples of rules: Rule 1: IF ANN1 is low AND ANN2 is low THEN Performance is very low. Rule 2: IF ANN1 is low AND ANN2 is medium THEN Performance is low. In Layer 5, the system's output is a crips number that represents player's performance value. After this joining process, player's final decision values were listed in descending order. This list has been compared with coach's list that was taken from the coach by using Spearman Rank Correlation Test. As a result of the comparisons, it has been seen that there is a strong relationship between developed the NFDSS result and coach's result. As the obtained results, it is seen that developed model is appropriate and consistent. Keywords: Neuro-Fuzzy systems, decision support systems, artificial neural networks, fuzzy logic, multi-criteria decision making.