Imputation Techniques Analysis for Incomplete Medical Datasets in Case-Based Reasoning System

Case based reasoning (CBR) based expert systems are widely accepted for healthcare-related applications. CBR is analogous to how humans solve problems using prior experience. CBR is a lazy learner and maintains the cases in case-base to solve the encountered target case. The accuracy of any case based reasoning based system depends upon the quality of cases in its case base. Nowadays, an ample amount of electronic healthcare data is available, but unfortunately, electronic data come along with a growing amount of missing data. Multiple techniques are available to handle those missing values. One widely used way is to use imputation methods to compute missing values for different attributes, which utilize the total actual values in the given dataset. The study was conducted to find the best suited imputation technique for calculating missing values for constructing case base in case-based reasoning system. This study uses benchmark medical dataset to examine four combination imputation techniques and data preprocessing method. The author proposes a hybrid model that performs the pearson correlation to find the critical feature and then used imputation techniques to fill in missing values in different attributes. Experiment results show that the Pearson correlation and linear regression (LR) outperform the other 3 imputed techniques' and give more accurate imputed values. To justify the implication of data preprocessing methods with imputation technique, the findings are carried to medical dataset collected from UCI.