CN-GAIN: Classification and NormalizationDenormalization-Based Generative Adversarial Imputation Network for Missing SMES Data Imputation

Quality data is crucial for supporting the management and development of SMES carried out by the government. However, the inability of SMES actors to provide complete data often results in incomplete dataset. Missing values present a significant challenge to producing quality data. To address this, missing data imputation methods are essential for improving the accuracy of data analysis. The Generative Adversarial Imputation Network (GAIN) is a machine learning method used for imputing missing data, where data preprocessing plays an important role. This study proposes a new model for missing data imputation called the Classification and Normalization-Denormalization-based Generative Adversarial Imputation Network (CN-GAIN). The study simulates different patterns of missing values, specifically MAR (Missing at Random), MCAR (Missing Completely at Random), and MNAR (Missing Not at Random). For comparison, each missing value pattern is processed using both the CN-GAIN and the base GAIN methods. The results demonstrate that the CN-GAIN model outperforms GAIN in predicting missing values. The CN-GAIN model achieves an accuracy of 0.0801% for the MCAR category and shows a lower error rate (RMSE) of 48.78% for the MNAR category. The mean error (MSE) for the MAR category is 99.60%, while the deviation (MAE) for the MNAR category is 70%.