Air Pollution Monitoring in Intelligent Cities Using Weighted Association Rule Mining

Introduction

With the emergence of newer technologies in recent years, various types of electronic data are being collected via sensors and smart devices, and analyzed through the use of intelligent data mining systems, to ensure that assets, resources and processes are managed more efficiently. Environmental science is one of the relevant areas that use intelligent systems to produce solutions for environmental problems and to protect the environment against factors such as pollution, adverse climate change and environmental degradation. In this context, observing air pollution data and analyzing and processing them are among the useful studies in this field.

In the current study, it is aimed to design an intelligent environment system that provides air quality control in cities by processing air pollution data collected from IoT sensors attached to smart devices in monitoring stations. For this purpose, Association Rule Mining (ARM) process is applied, which is one of the popular data mining methods. According to the rules formed by different combinations of the specified features of the available dataset, normal conditions are determined in terms of air quality factors to establish an intelligent system to detect the critical and exceptional situations. One drawback of the traditional ARM algorithms is that each item or factor is assumed to have the same significance even though some of them have greater importance compared to others within a transaction or within the whole item space. Therefore, the setting of the traditional ARM models should be organized in a weighted manner according to importance levels. In this respect, Tao et. al. (2003) developed an improved version of ARM, which is called Weighted Association Rule Mining (WARM). In this study, because different air pollutants/meteorological factors have different effects on air pollution, the rules are generated using WARM by taking account of item weights.

In the literature, several WARM models with different properties have been proposed. Wang et. al. (2000) introduced an algorithm, called WAR (Weighted Association Rules), which consists of two main parts. In the first part, the frequent item sets are generated through traditional ARM algorithms by ignoring the weights. Then post-processing is applied on the frequent item sets during rule-generation by introducing the intervals of item weight. WUARM (Weighted Utility ARM) developed by Khan et. al. (2008) considered both the significance of items and their utility as frequency of occurrences in transactions. Some of the WARM models were implemented using frequent pattern growth algorithm such as WIP (Weighted Interesting Pattern) mining and WFIM (Weighted Frequent Itemset Mining) as reported by Yun and Leggett (2005) and Yun and Leggett (2006). While WFIM was applied by creating ascending weight ordered prefix tree, WIP used a new concept of a weighted hyperclique pattern which considered weight confidence to discover frequent patterns with similar levels of weights and the h-confidence to determine affinity patterns with strong support. The proposed model used in this chapter was implemented using Apriori approach because pattern growth algorithms take huge amount of memory while storing in tree structure, which may not fit in memory when data size is very large.

A number of studies have focused on fuzzy set theory combined with WARM to mine weighted boolean and quantitative data (Wang and Zhang, 2003; Yue et. al., 2000; Lu and Qian, 1999; Gyenesei, 2000; Muyeba et. al., 2008). While FWARM (Fuzzy Weighted Association Rule Mining) proposed by Muyeba et. al. (2008) resulted in a valid downward closure property, others could not guarantee the validation. However, fuzzy algorithms have several disadvantages such as requiring knowledge about the problem solved and difficulties in setting the rules and choosing membership functions. Our algorithm differs from many other reported in the literature in terms of easy implementation and application.