A Simple Classifier Based on a Single Attribute

Seeking a simple but effective classifier is exciting and meaningful in both machine leaning and data mining. As usual, simplicity and high performance are two sides of a same coin. Our aim is to explore an easy-to-use classifier without losing its effectiveness. On this account, a single attribute based classification (SAC) algorithm is proposed. SAC first splits the original data set into multi one-dimensional data sets. After that, it creates a base classifier, e.g. C4.5, for each one-dimensional data, and then selects all classifiers having the highest accuracy. At last, SAC uses these selected classifiers to make prediction and the most frequent label is assigned to the new instance. Results of classification accuracy on 16 data sets from UCI machine learning repository indicate that the proposed method performs better in comparison with classical OneR algorithm. Experiments on high-dimensional data are also conducted to evaluate the proposed method, which demonstrates its scalability.