<![CDATA[Managing missing data is a critical aspect of preprocessing in data mining endeavors, significantly influencing output accuracy during both model development and utilization phases. This study introduces a novel approach to predicting missing values by partitioning data into disjoint subsets based on partitioning measures. The rationale behind this approach is the elimination of unrelated data through partitioning, thereby improving the accuracy of missing value prediction within each subset. Through a combination of expert panel insights and statistical tests (including the Chi-square test and Cramer's V coefficient), the database partitioning measure was determined using operational data from the Mashhad Fire and Safe Services Organization. Models were constructed for each partition, and missing data were segmented accordingly, with the corresponding models employed for prediction. The results revealed that in 44% of cases, models built on partitioned data outperformed those constructed on the entire dataset. The evaluation of this method underscores its capability to predict missing values with heightened accuracy. Notably, this approach is independent of the method employed for missing value prediction, enabling seamless integration into existing methods as an additional step to bolster prediction accuracy. ]]>
Copyrights © 2024
