EVALUATION OF IMPUTATION TECHNIQUES ON FUEL EMISSON FACTOR DATA

Sammanfattning

The globally increasing focus on environmental accountability has emphasized the need for accurate data and reliable forecasting to support informed sustainability efforts. However, incomplete datasets limit the capability of precise predictions. This study aims to address this issue through the application of imputation techniques on a fuel emission factor dataset. A selection of imputation methods was tested to impute missing values, and evaluated using error metrics. The results showed that Multiple Imputation by Chained Equation in combination with K-Nearest Neighbor or Random Forest are promising solutions for data imputation. Although, the imputation was not as accurate as anticipated and did not provide reliable enough values to reflect the dataset. The findings highlight the importance of selecting an imputation method suitable for the specific usecase together with qualitative, quantitative and representative data. Future work suggest testing additional methods, such as ensemble or hybrid models and dimensionality reduction, as well as incorporating a larger dataset with additional features.