This thesis aims to explore different approaches to handling missing values for Exploratory Data Analysis. This is done to illustrate the importance of picking a suitable approach for any giving dataset. The motivation for this stems from the importance of showing an accurate representation of data. Small deviations in the result can have major implications depending on the subject regarding the dataset. To answer this question this thesis shows the process of an EDA, and it will document the differences in results between these approaches. Data normalization (if applicable) and the handling of empty datapoints are explored and we found that the approach can have a significant impact on the result. How significant the impact is varies depending on the number of missing values and approach. With a larger number of missing values, the impact is very significant compared to when there are fewer missing values since more data changes. We can conclude that the approach does matter but it is equally important to consider the number of missing values in the dataset and where these values are located.
The subject for the EDA in question is Co2 emission rates per country and the dataset used contains 61,425 rows and 12 columns. The size of the dataset is important since too small a sample size could potentially show inaccurate results if that sample contains outliers. The analysis used the programming language Python to conduct all handling and visualization of data and Jupyter Notebook for structuring of code.