Initial Seed Value Effectiveness on Performances of Data Mining Algorithms

Abstract

After 2000s, Computer capacities and features are increased and access to data made easy. However, the produced and recorded data should be meaningful. Transformation of unprocessed data into meaningful information can be done with the help of data mining. In this study, classification methods from data mining applications are studied. First, the parameters that make the results of the same data set different were investigated on 4 different data mining tools (Weka, Rapid Miner, Knime, Orange), It has been tested with 3 different algorithms (K nearest neighborhood, Naive Bayes, Random Forest). In order to evaluate the performance of the data set while creating the classification models, the data set was divided into training data and test data as 80% -20%, 70% -30% and 60-40%. The accuracy, roc and precision values was used to test the performance of the classifying data. While classifying, the effect of algorithm parameters on the results is observed. The most important of these parameters is the initial seed value. The initial seed is a value using especially in classification algorithms that determines the initial placement of the data and directly affects the result. In this respect, it is very important to determine the initial seed value correctly. In this study, initial seed values between 0 and 100 were evaluated and it was shown that the classification could change the accuracy value approximately by 5%.