International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net AIR POLLUTION DETECTION USING MODIFIED TRAINGULAR MUTATION BASED PARTICLE SWARM OPTIMIZATION Pawandeep Kaur1, Prabhdeep Singh2, Kirandeep Singh3 1,2,3Department of Computer Science & Engineering, IKGPTU ----------------------------------------------------------------------------***-------------------------------------------------------------------------- ABSTRACT - The increase in population, emissions from industries and manufacturing activities, automobiles exhaust, etc., are reasons that are contributing to the air pollution. Many countries have declared it as major threat to human life. Currently air pollution is measured by utilizing spatially distributed sensors. However, due to sensor expenses and size limits the operational efficiency. Therefore, many researchers have proposed air pollution detection system using machine learning tools without deploying any particular kind of sensors. It reduces the cost of air pollution monitoring system. In this research work, a novel particle swarm optimization-based machine learning models to predict the concentration of benzene in the air has been designed to overcome the gaps found in the literature. Additionally, to modify particle swarm optimization-based machine learning models by considering the triangular mutation operator. It has improved the global and local search capabilities of particle swarm optimization and also it increases the convergence speed. Extensive analysis reveals that the proposed technique outperforms the existing techniques. KEYWORDS: AIR POLLUTION, BENZENE, PREDICTION, MACHINE LEARNING 1. INTRODUCTION Earth’s climate is changing rapidly and there are good reasons to believe that the release of carbon emissions by human activity is one of the many causes. This climate change comes in many forms such as variations in precipitation patterns and variations in storm patterns. One of the lesser studied forms is the change in wind patterns, both at low and high altitudes [1]. This project will look at global carbon emissions, global temperature anomalies, and global wind pattern anomalies at high altitude to see if there are geographic and / or temporal correlations between the three metrics [2]. The sources responsible for air pollution are of two categories which are natural sources and man-made sources. The natural sources include forest fires, volcanic eruption, and wind erosion of soil, natural radio activity and decomposition of organic matter by bacteria [3]. The manmade sources are much diversified. These include automobile, industries, thermal power plants and agricultural activities [4]. The fossils fuels (coal, oil, natural gas) are burnt in industries, thermal power plants and automobiles. Due to this carbon monoxide (C0), carbon dioxide (CO2), sulphur dioxide (SO2), sulphur trioxide (SO3) and nitrogen oxides are emitted. Different hydrocarbons (methane, butane, ethylene, benzene) and suspended particulate matters (dust, lead cadmium, chromium, arsenic salt etc.) are also present in these emissions [5]. These gases and suspended particulate matter (SPM) produced as result of burning fossils fuels are the greatest source of air pollution [6]. The pollutants released from natural sources of air pollution are dispersed in a vast area and do not cause any serious damage. Most of the health-related air pollutants come from man-made sources of air pollution [7]. In large cities, breathing the polluted air proves harmful to human health. Carbon monoxide, a serious air pollutant, reduces the oxygen carrying capacity of blood and causes nausea, headache, muscular weakness and slurring of speed. Oxides of nitrogen can damage the lungs, heart and kidneys of man and other creatures [8]. The presence of hydrocarbon in air causes irritation to eyes, bronchial construction, sneezing and coughing. In densely populated cities, the air pollution may take the form of industrial smog and photo chemical smog [9]. People are paying increasing attention to air pollution over the past decades, since it significantly impacts human health and causes serious harm to other living organisms such as animals and food crops. A lot of air pollution monitoring stations have been built around the world to inform people about air pollution status, such as the concentration of PM2.5, PM10, and O3. Besides monitoring, there is an urgent demand for air pollution prediction, which could benefit the government’s policy-making and public outdoor activities. The detailed classification of air pollutants by sources and usage of fuel is given in the Table 1. © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2005 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net Table 1: Principal sources of urban air pollutants Sr. No Sources Fuel Air Pollutants 1 Domestic heating and Cooking Wood, biomass etc Coal Light oil, kerosene, LPG SPM, carbon monoxide, oxides of nitrogen SPM, sulphur dioxide, carbon monoxide, oxides of nitrogen SPM, sulphur dioxide, carbon dioxide, oxides of nitrogen 2 Industrial boilers and power plants Coal, light diesel oil, furnace oil, heavy oil SPM, sulphur dioxide, carbon dioxide, oxides of nitrogen 3 Industrial manufacturing Processes -- Iron oxides from steel industry, dust/PM from cement industry etc. 4 Transportation Gasoline, LPG, CNG Sulphur dioxide, carbon monoxide, oxides of nitrogen, hydrocarbons, oxidants, ozone and lead 5 Transportation/vehicle etc. Petrol/diesel SPM, sulphur dioxide, odour, oxides of nitrogen, carbon dioxide 6 Community service such as municipal incinerators, biomedical incinerator, fossil fuel burning etc SPM, sulphur dioxide, oxides of nitrogen, carbon monoxide, volatile organic compounds, lead The major air pollutants, their sources and health effects are summarized in Table 2. Table 2: Major air pollutants, their sources and health effects Sources Health Impact Carbon dioxide (CO2) Burning of fossil fuels(coal, oil etc) in furnaces Respiratory problems, green house effect, global warming. Carbon monoxide (CO) Automobile, industrial furnaces, open fires, forest fires and combustion of domestic fuel. Difficulty in breathing, headache and irritation of mucous membranes and death. Sulphur dioxide (SO2) Burning of fossil fuels, industries and automobile. Burning of fossil fuels, industries and automobile. death. Hydrogen sulphide (H2S) Decaying vegetation and animal matter. Sulphur springs, volcanic eruptions and sewage treatment plants etc. Irritation of alveoli, lung and respiratory infections and death. Nitrogen oxides (NOx) Industries manufacturing HNO3 and other chemicals and automobile exhaust. Irritation of alveoli, lung and respiratory infections and death. Air Pollutants © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2006 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net Table 3 depicted the important sources of major ambient air pollutants. Table 3: Sources of air pollution Pollution Formation Physical state Sources Particulate Primary Solid, liquid Construction sites, unpaved roads, fields, smokestacks, volcanoes, dust storms, forest and grassland fires, living vegetation, and sea spray emitted from power plants, industries and automobiles Matter Secondary S Oxides of Nitrogen Primary Gas Vehicles, gas heaters and cookers Oxides of Carbon Primary Gas Vehicles and tobacco smoke Sulfur dioxides(SO2) Primary Gas Industrial processes and vehicles Volatile organic compounds(VOCs) Primary and secondary Gas Vehicles, smoke Lead (Pb) Primary Solid (fine particulates) Vehicles, industries Photochemical oxides such Ozone (O3) Secondary Gas Secondary to photo-oxidation of NOx and VOCs and industries, tobacco Table 4: Revised national ambient air quality standards for major pollutants, India Pollutants Time-weighted average Concentration in ambient air Industrial Areas Residential, Rural & other Areas Ecologically Sensitive Areas SO2 Annual 24 hours 50 80 20 80 NOx Annual 24 hours 40 80 30 80 PM10 Annual 24 hours 60 100 60 100 PM2.5 Annual 24 hours 40 60 40 60 Ozone(O3) 8 hours 1 hours 100 180 100 180 Lead (Pb) Annual 24 hours 0.50 1.0 0.50 1.0 Ammonia1 Annual 100 100 (NH3) 24 hours 400 400 CO 8 hours 1 hour 2 4 2 4 © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2007 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net 2. RELATED WORK Dominici et al. (2010)[7] Modeling the joint behavior of multivariate extreme events is of interest in a wide range of applications, ranging from telecommunications to Earth and environmental sciences, such as hydrology, seismology, and applications related to climate change or air pollution monitoring. Simultaneous exposure to multiple air pollutants seriously affects public health worldwide, causing loss of life and livelihood and requiring costly health care. Therefore, policymakers such as those at the US Environmental Protection Agency (EPA), are researching multivariate approaches to quantify air pollution risks the issue of air pollution is compounded by global warming and climate change, as increasingly high temperatures are suspected to contribute to raising ozone concentrations and aggravating their ect in the human body Kahle et al. (2015)[8]. This pressing situation urges a greater understanding and better monitoring of air pollution extremes, the complexity of which poses a challenge for standard statistical techniques. Indeed, multiple air pollutants are often recorded at various spatial locations and, therefore, the modeling of peak exposures across a spatial region must transcend the assumption of independence in order to capture their spatial variability. In this paper, we propose a new methodological framework based on Extreme-Value Theory, for estimating the probability that multiple air pollutants and temperatures will be simultaneously extreme at different spatial locations. The statistical modeling of single extreme variables observed over space is usually based on spatial max-stable processes. Davison et al. (2012)[9]which are the only possible limit models for properly renormalized block maxima and whose block size increases to in nity. Except in the case of asymptotic independence, these models can be used to capture the potentially strong spatial dependence that may exist among variables at extreme levels. Models for asymptotically independent data in the bivariate case were introduced by Huser et al. (2017)[10] proposed several spatial models for high threshold exceedances that can handle both asymptotic dependence and independence; see also the related paper by Krupskii et al. (2018)[11]. However, the current literature on multivariate modeling of spatial extremes is still rather sparse. Here, we restrict ourselves to asymptotic dependence by modeling multivariate block maxima recorded over space using a suitable max-stable process. Oesting et al. (2017)[12] introduced a bi variant Brown{ Resnick max-stable process to jointly model the spatial observations and forecasts of wind gusts in Northern Germany. In this paper, we propose a new class of multivariate maxstable processes that extends the Reich{ Shaby model Reich et al.(2012)[13] to the multivariate setting, and that is suitable for studying the spatial and cross-dependence structures of multiple max-stable random fields within an intuitive and computationally convenient hierarchical tree-based framework. In contrast to the standard spatial processes based on the Gaussian distribution, the computationally demanding nature of the likelihood function for max-stable processes has hampered their use in high-dimensional settings within both frequentist and Bayesian frameworks Thibaud et al. (2016)[14] showed how the Brown{Resnick max-stable process may be submitted using a well-designed Markov chain Monte Carlo (MCMC) algorithm; however, this remains excessively expensive in high dimensions. From a computational perspective, it is convenient to relax the max-stable structure by assuming conditional independence of the extreme observations given an unobserved latent process Opitz et al., 2018)[15]. This significantly facilitates Bayesian and likelihood-based inference and may be helpful for estimating marginal distributions by borrowing strength across locations. Unfortunately, when the latent process is Gaussian, the resulting dependence structure lacks exibility and cannot capture strong external dependence. Reich and Shaby, 2012)[16] on the other hand possesses a conditional independence representation given some latent stable random effects and is jointly max-stable. Other popular max-stable processes do not possess such a convenient hierarchical characterization. In this paper, we generalize the Reich {Shaby process for the modeling of multivariate spatial extremes by assuming a nested, tree-based,-stable latent structure, which allows us to maintain a moderate computational burden. In our proposed model, each variable is described using a Reich Shaby spatial process and the dependence among © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2008 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net variables at different locations may be expressed in terms often asymmetric max-mixture of multivariate nested logistic distributions Stephenson, 2003)[17]. The global dependence structure may be represented by a tree framework, in which the\leaves"(i.e., the terminal nodes) correspond to the different spatial max-stable processes representing different variables of interest (e.g., pollutants), and the tree \branches" describe the relationships among these processes, which are grouped into clusters. The intra-cluster cross-dependence is assumed to be exchangeable and stronger than the inter-cluster cross dependence .In principle, the underlying tree structure might involve an arbitrary number of\layers" (i.e., node levels) in order to describe more complicated forms of cross-dependence among the spatial processes, although more complex trees necessarily imply an increased number of latent variables and parameters, thus complicating the inference procedure. The remainder of this paper is organized as follows. The theoretical background for the geostatistical modeling of extremes is reviewed in x2. In x3, we introduce a novel class of hierarchical max-stable processes built from nested -stable random effects for the joint modeling of multivariate extremes over space, and study its dependence properties. We3also detail our inference procedure based on an MCMC algorithm, and describe some simulation experiments. In x4, we use our nested max-stable model to investigate the spatial and cross-dependence structures of the concentration maxima of various air pollutants and temperature, observed at a number of sites across the Los Angeles area in California, US. x5 concludes with some remarks and perspectives for future research. Chang et al. (2011)[18]. The World Health Organization classified air pollution as a major environmental health risk, estimating that 4.2 million premature deaths in can be attributed to exposure to outdoor air pollution (World Health Organization and others,. Nitrogen dioxide (NO2), a highly reactive gas monitored by the EPA, is formed primarily from fuelburning emissions. Katsouyanni et al. (2009)[19]. Health effects from short-term exposures include cardiovascular effects and premature mortality as well as difficulty breathing and increased occurrence of hospital visits due to decreased lung capacity including asthma exacerbation (US Environmental Protection Agency, McKenzie et al. (2016)[20]. Long term effects include cardiovascular effects, premature mortality, diabetes, poorer birth outcomes, cancer, and asthma in children (US Environmental Protection Agency, 2016). NO2 also reacts with water in the atmosphere to produce ozone and acid rain. The resulting nitrate particles from this reaction can contribute to particulate matter less than 2.5 microns in diameter all of which have additional environmental health considerations. NO2 can react with water, ozone, and nitric oxide (NO) multiple times over a span of several hours to form and re-form NO2 and NO. Emissions of nitroous oxides are primarily in the form of NO, where 92% of NO is anthropogenic with 56% estimated to be from mobile emissions. Kim et al.(2012)[21]Hence understanding patterns of mobile emissions are paramount to protecting public health, particularly for susceptible populations, such as children, the elderly, and those with asthma or compromised immune systems. The vast majority of health studies to date have focused on the relationship between human health effects and longer term exposures, such as 1-hour or daily average aggregate effects. Additionally, these studies are based on air-quality measurements for which spatial information is often limited due to the number of stationary monitors measuring air quality over large regions. Adler et al. (2015)[22]. International efforts include Air Map Korea Project (Rooney, 2018), which aims to install over 4.5 million monitors on telephone poles, public phone booths and central offices, and Smart City Barcelona’s Lighting Masterplan (Adler, 2016), which equipped lampposts with air quality sensors to relay information to the city and to the public. London showed creativity in air quality monitoring with their Pigeon Patrol McKenzie et al. (2016)[23] fitting pigeons with mobile air quality sensors to measure nitrogen dioxide across the city. Methodology utilizing this information to provide real-time air pollution maps as well as short-term air quality forecasts on a fine-resolution temporal and spatial scale may revolutionize people’s understanding of their personal environment and exposures, having real-world implications and impacts on citizens. As the effects of air pollution and the differences in pollutant microenvironments become more widely studied and understood, people are increasingly concerned with understanding their immediate personal environment and its effect on their health. They are interested in real-time information on a very localized scale in order to make informed decisions on their day-to-day activities. Schiffman et al. (2017)[24] EPA’s widely used Air Quality Index (AQI) uses high quality hourly data from stationary Federal Reference Monitors (FRMs) and Federal Equivalency Monitors (FEM) to implement a color-coded air quality scale and public © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2009 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net messaging system. Real-time and forecasted AQI’s are provided to the public via EPA’s AirNow website (airnow.gov) and are available by city, state, or ZIP code. Localized sensor networks, which may include mobile sensors, may provide additional information on which to build, refine, and expand air quality information available to the public. Apte et al. ( 2017)[25]. The cars drive through the city on spatiotemporal tracks and measure ambient NO2 each second. This study provides a unique source of highly detailed data with spatial band temporal complexities. It can provide information about commuter exposure, hot spots within high-trafficked city streets, as well as complex patterns due to meteorological effects and microenvironments. This fine-scale spatial and temporal information could also lead to the methodology and information needed to start to characterize acute exposure. It is particularly important to understand near-road and citystreet environments. The US Environmental Protection Agency reports that over 45 million people live in close proximity to major roadway (US Environmental Protection Agency, 2018). The novel data set provides information about air quality surrounding roadways and commonly trafficked areas that is not available from the limited number of stationary monitors across an area or region. Snyder et al. (2013)[26] Fine-scale air quality measurement and analysis have been powered by recent advances in sensor technologies, which allow for the use of mobile network platform with low-cost sensors for the purpose of general monitoring As a complement to fixed point networks, mobile air quality monitoring can improve spatial coverage and be used to. Dutta et al.(2009)[27] smart-phone map air pollution with improved spatial and temporal resolution. In particular, finescale air quality monitoring is essential in urban settings, because the measurements vary dramatically over space and time and closely relate to several factors, such as land use, traffic, and meteorology. Several mobile platform have been developed, such as wearable device. Finally, we conduct a simulated experiment to determine the relative effectiveness of a fleet of mobile monitors compared to a network of fixed-location monitors, and find that mobile monitors provide more accurate estimation and prediction compared to limited fixed-location monitors. 3. PROPOSED METHODOLOGY This section contains graphical flowchart and step by step methodology of the proposed technique. Figure 5: Graphical representation of proposed methodology © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2010 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net Given an input sequence a standard recurrent neural network (RNN) computes the hidden vector sequence and output vector sequence by iterating the following equations from where the terms denote weight matrices (e.g., is the input-hidden weight matrix), the terms denote bias vectors (e.g., is hidden bias vector) and is a hidden layer function. is usually an element-wise application of a sigmoid function. A conventional RNN often confronts poor correctness when executing long sequences due to the explosion gradient problem. Replacing a powerless logistic function with Long Short-Term Memory (LSTM) units can solve this issue. Although many LSTM architectures differ in their structure and activation functions, all of them have explicit memory cells with complicated dynamics allowing it to easily “memorize” information for an extended number of timesteps. The LSTM architecture used in our experiments is given by the following equations: where σ is the logistic sigmoid function, and , , , and are respectively the input gate, forget gate, output gate, and cell activation vectors, all of which are the same size as the hidden vector . The weight matrices from the cell to gate vectors (e.g., ) are diagonal, so element m in each gate vector only receives input from element m of the cell vector. Encoder-Decoder model In the model, an encoder reads the sequence of vectors mentioned in section 2.1. into a vector using RNN with LSTM units as where is a hidden state at time and is a vector generated from a sequence of hidden vectors, and nonlinear functions. This model uses an RNN with LSTM units as and a mean operation as . are some A decoder is often trained to predict the value ′ given the context vector and all the previous predicted values In other words, a conventional decoder defines a probability over the prediction by decomposing the joint probability into the conditional order: ∏ where ( ). With an RNN, each conditional probability is modeled as follows: where is a nonlinear, potentially multi-layered, function that outputs the probability of , and is the hidden state of the RNN. Encoder-Decoder model can stack multiple RNN layers on the top of each other to increase correctness. © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2011 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net 4. PERFORMANCE ANALYSIS The presented method was implemented on two datasets collected from multiple resources. More specially, weather data from July 2008 to April 2018 of researched cities were crawled from World Weather Online and Time & Date while AQI information was collected from Air Korea, Seoul Clean Air, Chinese PM2.5 report, and Daegu Environment Government private APIs. Next, Spark was used to clean irrelevant data, fulfill missing elements, join data from various resources to one jointly dataset, and transform context features into vectors, in which, each vector corresponds to one-hour timestep. In particular, a rough dataset of Daegu is approximately 10Gb with more than 30 million records. After pre-processing, the analyzing engine kept only 27.936 rows. The dataset was separated into two parts, in which, the first part related to the period from January 2015 to June 2017 due to the missing of air quality information, and the second one was from June 2017 to March 2018. At first, we trained the model on the first part then transfers the learned weights to the second one instead of training from scratch. The period from February to March 10th, 2018 was used as the testing set. On the contrary, Seoul data included complete information from January 2008 to April 2018 with more than 2 million records corresponding to hourly records of 25 districts. Additionally, on Seoul dataset, the model did not use transfer method; data from 2008 to 2016 was used as training set, and test set was the data from January 2017 to April 2018. Both datasets were randomly carved out twenty percent of the training set to be the validation set. © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2012 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net Fig. 2. Shows Monthly Distribution of PM2.5 AQI in Seoul (2008-2018). (a) Shows only the aggregation of the Unhealthy levels when PM2.5 AQI is above 100 and 150, (b) reveals all four warning level distribution. As shown in Figure 3, dangerous categories increase substantially from October of a current year to March of the next year and reach the peak point in January. After that, it gradually declines from March to the lowest point in September of the next year. Consequently, months in a year, hours in a day, holidays information were also elements of featured vectors. WPSO WITH PSO MPSO 97.2656 91.9271 99.4792 97.3 91.6 99.5 Root Mean Square 0.1796 0.2049 0.1449 Error 0.973 0.919 0.995 Tp rate 0.039 0.149 0.006 Precision 0.973 0.927 0.995 Kappa Statistic (Ks ) 0.9395 0.8126 0.9885 Accuracy F-Measures © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2013 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net 5. CONCLUSION Many countries have declared air pollution as a major threat to human life as it causes several diseases. Currently air pollution is measured by utilizing spatially distributed sensors, but due to sensor expenses and range limits the operational efficiency. Therefore, many researchers have proposed air pollution detection system using machine learning tools without deploying any particular kind of sensors. Thus, it reduces the cost of air pollution monitoring system. The reason behind the study is to predict the concentration of benzene, which is considered as most hazardous for human being as it causes blood cancer to individuals. In this research work, a novel particle swarm optimization-based machine learning models to predict the concentration of benzene in the air has been designed to overcome the gaps found in the literature. Additionally, to modify particle swarm optimization-based machine learning models by considering the triangular mutation operator. It has improved the global and local search capabilities of particle swarm optimization and also it increases the convergence speed. Extensive analysis reveal that the proposed technique outperforms the existing techniques. REFERENCES [1] Vardoulakis S, Fisher BEA, Pericleous K, Gonzalez-Flesca N. Modelling air qualityin street canyons: a review. Atmospheric Environment 2003; 37:155-82. [2] Donnelly A, Misstear B, Broderick B. Real time air quality forecasting using integrated parametric and non-parametric regression techniques. Atmospheric Environment 2015;103:53-65. [3] Zheng Y, Yi X, Li M, Li R, Shan Z, Chang E, et al. Forecasting Fine-Grained Air Quality Based on Big Data. KDD 2015. Sydney, NSW, Australia2015. [4] Liang Y, Ke S, Zhang J, Yi X, Zheng Y. GeoMAN: Multi-level Attention Networks for Geo-sensory Time Series Prediction. IJCAI-182018. p. 3428-34. [5] Z. Qi and et al., “Deep Air Learning: Interpolation, Prediction, and Feature Analysis of Fine-grained Air Quality,” TKDE 2018, unpublished. [6] J. Zhang, D. Qi, “Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction,” AAAI 2017. [7] Dominici, F., Peng, R. D., Barr, C. D. and Bell, M. L. (2010) Protecting human health from air pollution: shifting from a single-pollutant to a multi-pollutant approach. Epidemiology 21(2), 187{194. [8] Kahle, J. J., Neas, L. M., Devlin, R. B., Case, M. W., Schmitt, M. T., Madden, M. C. andDiaz-Sanchez, D. (2015) Interaction e_ects of temperature and ozone on lung functionand markers of systemic inammation, coagulation, and _brinolysis: A crossover study ofhealthy young volunteers. Environ Health Perspect 123(4),310{316. [9] Davison, A. C. and Gholamrezaee, M. M. (2012) Geostatistics of extremes. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 468, 581{608. [10] Huser, R., Opitz, T. and Thibaud, E. (2017) Bridging asymptotic independence and dependence in spatial extremes using Gaussian scale mixtures. Spatial Statistics. 21, 166{186. [11] Krupskii, P., Huser, R. and Genton, M. G. (2018) Factor copula models for replicated spatial data. Journal of the American Statistical Association To appear. [12] Oesting, M., Schlather, M. and Friederichs, P. (2017) Statistical post-processing of forecasts for extremes using bi variate Brown {Resnick processes with an application to wind gusts. Extremes 20(2), 309{332. [13] Reich, B. J. and Shaby, B. A. (2012) A hierarchical max-stable spatial model for extreme precipitation. Annals of Applied Statistics 6(4), 1430{1451. [14] Thibaud, E., Aalto, J., Cooley, D. S., Davison, A. C. and Heikkinen, J. (2016) Bayesian inference for the Brown{Resnick process, with an application to extreme low temperatures. Annals of Applied Statistics 10(4), 2303{2324. © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2014 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 09 | Sep 2019 p-ISSN: 2395-0072 www.irjet.net [15] Opitz, T., Huser, R., Bakka, H. and Rue, H. (2018) INLA goes extreme: Bayesian tail regression for the estimation of high spatio-temporal quantiles. arXiv preprint 1802.01085. [16] Reich, B. J. and Shaby, B. A. (2012) A hierarchical max-stable spatial model for extreme precipitation. Annals of Applied Statistics 6(4), 1430{1451. [17] Stephenson, A. (2003) Simulating multivariate extreme value distributions of logistic type. Extremes 6(1), 49{59. [18] Chang, H. H., Reich, B. J. and Miranda, M. L. (2011) Time-to-event analysis of fine particle air pollution and preterm birth: results from North Carolina, 2001–2005. American Journal of Epidemiology, 175, 91–98. [19] Katsouyanni, K., Samet, J. M., Anderson, H. R., Atkinson, R., Le, A. T., Medina, S., Samoli, E., Touloumi, G., Burnett, R. T., Krewski, D. et al. (2009) Air pollution and health: a European and North American approach (APHENA). Research Report (Health Effects Institute), 5–90. [20] McKenzie, S. (2016) Pigeon Air Patrol to the rescue! Birds with backpacks track air pollution. CNN. Retrieved from: https://www.cnn.com/2016/03/16/europe/ pigeon-air-patrol-pollution-london/index.html [Last accessed on 2018 Jul 11]. [21] Kim, J. W., Jeong, Y. K. and Lee, I. W. (2012) Automatic sensor arrangement system for building energy and environmental management. Energy Procedia, 14, 265–270. [22] Adler, L. (2015) Monitoring air quality and the impacts of pollution. Data-Smart City Solutions. Retrieved from: https://datasmart.ash.harvard.edu/news/article/monitoring-air-quality-and-the-impacts-of-pollution-679 [Last accessed on 2018 Jul 11]. [23] McKenzie, S. (2016) Pigeon Air Patrol to the rescue! Birds with backpacks track air pollution. CNN. Retrieved from: https://www.cnn.com/2016/03/16/europe/ pigeon-air-patrol-pollution-london/index.html [Last accessed on 2018 Jul 11]. [24] Schiffman, R. (2017) A cheap pollution sensor will keep you off the dirtiest roads. NewScientist. Retrieved from: https://www.newscientist.com/article/ 2148503-a-cheap-pollution-sensor-will-keep-you-off-the-dirtiest-roads/ [Last accessed on 2018 Jul 11]. [25] Apte, J. S., Messier, K. P., Gani, S., Brauer, M., Kirchstetter, T. W., Lunden, M. M., Marshall, J. D., Portier, C. J., Vermeulen, R. C. and Hamburg, S. P. (2017) High-resolution air pollution mapping with google street view cars: exploiting big data. Environmental Science& Technology,51, 6999–7008. [26] Snyder, E. G., Watkins, T. H., Solomon, P. A., Thoma, E. D., Williams, R. W., Hagler, G. S., Shelow, D., Hindin, D. A., Kilaru, V. J. and Preuss, P. W. (2013) The changing paradigm of air pollution monitoring. [27] Dutta, P., Aoki, P. M., Kumar, N., Mainwaring, A., Myers, C.,Willett,W. andWoodruff, A. (2009) Common sense: participatory urban sensing using a network of handheld air quality monitors. In Proceedings of the 7th ACM conference on embedded networked sensor systems, 349–350.ACM. © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2015