This chapter explains the practically how to classify the oil spill data which identifies the similarities in the chemical properties using chemometric analysis. This chapter aimed to make a detailed understanding of the source and also behavior of spilled oil based on its chemical composition. Last but not least, this chapter emphasizes that accurate classification of oil spills requires rigorous data pre-processing to ensure high-quality analysis. Standardizing variables, handling outliers, and confirming normality are critical steps in producing meaningful and reproducible clustering results for environmental forensic investigation. The classification of oil spill dataset is imperative to determine the similarities of chemical properties within the cluster or dissimilarities with different clusters. It can be done using the chemometric analysis. However, prior to proceeding with the classification analysis, pre-treatment is another approach that must be successfully performed to enable the data properly center, scale, and reduce the dimensionality of the data. By doing this, it can improve predictive modeling and data quality, anomalies during the pre-treatment phase of oil spill data, while missing values are replaced with the mean (for continuous data) and median (for skewed data) techniques. Skewed data refers to the distribution of oil spill datasets if it is not symmetrical around the mean. For instance, the histogram or probability distribution has an uneven shape because it is “pulled” or “stretched” more to one side. Standardized skewness and standardized kurtosis are calculated for data verification that the data distribution is normal. Skewness quantifies how asymmetrical the data distribution is. It shows if the data points are distributed more to the left or right of the mean. Kurtosis provides insights about the shape of frequency distribution, data distribution's “tailedness.” It aids in comprehending the existence of outliers. In fact, both skewness and kurtosis as additional variables to see how they influence clustering. The data or variables’ goodness of fit to a normal distribution is assessed using the 95% confidence level normality tests, namely the Shapiro–Wilk, Anderson–Darling, Liliforths, and Jarque–Bera tests. The results of the tests will normally show that not all of the raw data is normally distributed (p < 0.05). The data is optimally transformed using log-transformation and z-scale normalized (mean = 0, standard deviation = 1). Standardization is crucial because the numerical data or parameters are subject to different unit ranges and measurements. Additionally, it makes it possible to minimize the impact of measurement discrepancies and guarantee that every variable has an equal effect. Data pre-treatment is essential because it guarantees that data is clear, consistent, and prepared for insightful analysis.

Classification of Similar Chemical Properties in Oil Spills

章节
发表日期：

2026年

查看全文

Azimah Ismail,
Hafizan Juahir

摘要

This chapter explains the practically how to classify the oil spill data which identifies the similarities in the chemical properties using chemometric analysis. This chapter aimed to make a detailed understanding of the source and also behavior of spilled oil based on its chemical composition. Last but not least, this chapter emphasizes that accurate classification of oil spills requires rigorous data pre-processing to ensure high-quality analysis. Standardizing variables, handling outliers, and confirming normality are critical steps in producing meaningful and reproducible clustering results for environmental forensic investigation. The classification of oil spill dataset is imperative to determine the similarities of chemical properties within the cluster or dissimilarities with different clusters. It can be done using the chemometric analysis. However, prior to proceeding with the classification analysis, pre-treatment is another approach that must be successfully performed to enable the data properly center, scale, and reduce the dimensionality of the data. By doing this, it can improve predictive modeling and data quality, anomalies during the pre-treatment phase of oil spill data, while missing values are replaced with the mean (for continuous data) and median (for skewed data) techniques. Skewed data refers to the distribution of oil spill datasets if it is not symmetrical around the mean. For instance, the histogram or probability distribution has an uneven shape because it is “pulled” or “stretched” more to one side. Standardized skewness and standardized kurtosis are calculated for data verification that the data distribution is normal. Skewness quantifies how asymmetrical the data distribution is. It shows if the data points are distributed more to the left or right of the mean. Kurtosis provides insights about the shape of frequency distribution, data distribution's “tailedness.” It aids in comprehending the existence of outliers. In fact, both skewness and kurtosis as additional variables to see how they influence clustering. The data or variables’ goodness of fit to a normal distribution is assessed using the 95% confidence level normality tests, namely the Shapiro–Wilk, Anderson–Darling, Liliforths, and Jarque–Bera tests. The results of the tests will normally show that not all of the raw data is normally distributed (p < 0.05). The data is optimally transformed using log-transformation and z-scale normalized (mean = 0, standard deviation = 1). Standardization is crucial because the numerical data or parameters are subject to different unit ranges and measurements. Additionally, it makes it possible to minimize the impact of measurement discrepancies and guarantee that every variable has an equal effect. Data pre-treatment is essential because it guarantees that data is clear, consistent, and prepared for insightful analysis.