Data Mining on Divers Alert Network DSL Database: Classification of Divers

Abstract

Divers Alert Network (DAN) created a database (DB) with a big amount of dive related data which has been collected since 1994 within the scope of Dive Safety Laboratory (DSL) project. The aim of this study is to analyze the DB using data mining techniques. The clustering of divers by their health and demographic information and reveal significant differences in diver groups are the main objectives of this study.

To eliminate time effect of age, divers who participated to only one dive were included in the study. The numbers of one-dive divers is 874. Before applying clustering methods, data cleaning was performed to eliminate the potential mistakes resulting from inconsistencies, inaccuracies and missing information. TwoStep, Gower distances and K-means clustering methods were performed on DB to find the naturally associated clusters. Conventional statistical analyses were performed to understand differences in clusters and between male and female divers.

As the result of these analyses, divers were separated into 3 clusters and distinguishing variables of these clusters were revealed. As TwoStep and Gower Distances are suitable for categorical variables, age and dive activity years were distributed in 3 categories. For K-Means Clustering, original numerical values of these variables was used. The most distinct clusters were formed by TwoStep Clustering. The middle aged male divers with without any health problem are in Cluster 1. Male and female divers with health problems and high rate of cigarette smoking are in the Cluster 2 and old divers with many dive activity years are in the Cluster 3. The search for significant differences in dive-related variables was performed based on the TwoStep Clustering results and separating male and female divers.