Mining Class Association Rules on Dataset with Missing Data

Many real-world datasets contain missing values, affecting the efficiency of many classification algorithms. However, this is an unavoidable error due to many reasons such as network problems, physical devices, etc. Some classification algorithms cannot work properly with incomplete dataset. Therefore, it is crucial to handle missing values. Imputation methods have been proven to be effective in handling missing data, thus, significantly improve classification accuracy. There are two types of imputation methods. Both have their pros and cons. Single imputation can lead to low accuracy while multiple imputation is time-consuming. One high-accuracy algorithm proposed in this paper is called Classification based on Association Rules (CARs). Classification based on CARs has been proven to yield higher accuracy compared to others. However, there is no investigation on how to mine CARs with incomplete datasets. The goal of this work is to develop an effective imputationmethod formining CARs on incomplete datasets. To show the impact of each imputation method, two cases of imputation will be applied and compared in experiments.