Leveraging Data Mining to Extract Accidental Drug Overdose Death Patterns: 2012-2014 US Dataset as Case Study

This study examined data on drug-related accidental deaths in the United States as found in the Accidental_Drug_Related_Deaths.csv dataset to understand patterns, trends, and risk factors and gain an understanding of the potential applicability of secondary data in public health planning. The dataset contains a total of 11,981 records with 48 single-value fields that include demographic information, location of event, and substances involved. The study undertook extensive data preprocessing on the dataset that included replacing missing values, standardizing elements of the dataset, reducing the data for analysis while maintaining the ability to examine the original structure, and transforming or restructuring fields for a meaningful analysis of the data. The study applied various data mining techniques such as association rule mining, classification, clustering, and outlier detection to draw insights from the dataset. The study identified high-risk demographic groups and combinations of drugs most often found in overdose situations, spatial hotspots for overdoses, and a few outliers. The study included several visualizations and interpretations of the data, and assessed ethical considerations of privacy, data exploitation or misappropriation, and biases. The study found data mining an effective data analysis strategy to help public health, policy development and emergency management organizations anticipate and/or mitigate drug overdose incidence and severity.