Empirical Comparison of Techniques for Handling Missing Values

The performance of all technologies is highly depended on the quality of the data. For example, Neural Network (NN) technique can be applied very well if the data have been
well prepared and free from noise and missing value. This study empirically compares several handling missing value methods for NN based on literature. Six of those methods have been identified and compared using adult data set (retrieved from UCI database). The methods include mean average, replace with one, replace with zero, replace with maximum, and replace with minimum and regression. The result shows that replace with maximum value method yield better accuracy compare to the other methods.