Performance Evaluation of Classification Algorithms on Different Data Sets

Objectives: The most appropriate classifier selections for the particular data sets were generally found harder. Therefore, in this study various existing classifiers have been considered on several data sets to assess their performance. Methods/ Statistical Analysis: Usually, the selections of classification techniques, such as, Naive Bayes (NB), Decision Tree (DT), Lazy Classifiers (LC), Support Vector Machine, etc., depend on the type and nature of the attributes in the data set. The wrong selection of classification technique can certainly lead to wrong results and poor performance. This concept is the motivation behind this study. Usually the data set consists of nominal attributes, numeric attributes or mix attributes (both numeric and nominal attribute). In this paper, different types of data sets are applied on three most popular classification techniques, such as, NB, DT, and LC, to evaluate their performances. Findings: The result reveals that NB classifier performs well on both mix attribute data and numeric data but decision tree classifier performs better on nominal attribute data. Lazy classifier’s performance is just average for all kind of data. Application/Improvements: The results of this study will helps in understanding the performance of different classification techniques on different data sets. Further, results can be utilized to select the best classification technique among NB, decision tree and lazy classifiers in order to use with different data sets.