**Regression Models to Predict Air Pollution from Affordable Data Collections Affordable Data Collections**

**Regression Models to Predict Air Pollution from** 

DOI: 10.5772/intechopen.71848

Yves Rybarczyk and Rasa Zalakeviciute Additional information is available at the end of the chapter

Yves Rybarczyk and Rasa Zalakeviciute

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/intechopen.71848

## **Abstract**

[7] Sim J, Park J-S, Kim M, Bae D, Choi Y, Kim L-S. A 1.42TOPS/W deep convolution neural network recognition processor for intelligent IoE systems. In: Proceeding of IEEE Inter-

[8] Bong K, Choi S, Kim C, Kang S, Kim Y, Yoo HJ. 14.6 A 0.62mW ultra-low-power convolutional-neural-network face-recognition processor and a CIS integrated with always-on haar-like face detector. IEEE International Solid-State Circuits Conference

[9] Desoli G et al. 4.1 A 2.9TOPS/W deep convolutional neural network SoC in FD-SOI 28nm for intelligent embedded systems. In: IEEE International Solid-State Circuits Conference

[10] Chen YH, Krishna T, Emer J, Sze V. 14.5 Eyeriss: An energy-efficient reconfigurable accelerator for deep convolution neural networks. In: IEEE International Solid-State Cir-

[11] Du L et al. A reconfigurable streaming deep convolutional neural network accelerator for internet of things. IEEE Transactions on Circuits and Systems I: Regular Papers. 2017; 99:1-11

[12] Han S, Mao H, Dally W. Deep compression: Compressing DNNs with pruning, trained

[13] Du Y et al. A streaming accelerator for deep convolutional neural networks with image and feature decomposition for resource-limited system applications. Sep. 2017. arXiv:1709.05116

[14] Du Y et al. A memristive neural network computing engine using CMOS-compatible

[15] Shin S, Kim K, Kang SM. Memristive computing- multiplication and correlation. In: IEEE International Symposium on Circuits and Systems; Seoul. 2012. pp. 1608-1611

[16] Desoli G et al. 14.1 A 2.9TOPS/W deep convolutional neural network SoC in FD-SOI 28nm for intelligent embedded systems. In: IEEE International Solid-State Circuits Con-

national Solid-State Circuits Conference (ISSCC). Jan/Feb. 2016. pp. 264-265

(ISSCC); San Francisco. 2017. pp. 248–249

14 Machine Learning - Advanced Techniques and Emerging Applications

(ISSCC); San Francisco. 2017. pp. 238–239

[cs.AR]

cuits Conference (ISSCC); San Francisco. 2016. pp. 262–263

quantization and huffman coding. 2015. arxiv:1510.00149v3

Charge-Trap-Transistor (CTT). Sep 2017. arXiv:1709.06614 [cs.ET]

ference (ISSCC); San Francisco. 2017. pp. 238-239

Air quality monitoring is key in assuring public health. However, the necessary equipment to accurately measure the criteria pollutants is expensive. Since the countries with more serious problems of air pollution are the less wealthy, this study proposes an affordable method based on machine learning to estimate the concentration of PM2.5. The capital city of Ecuador is used as case study. Several regression models are built from features of different levels of affordability. The first result shows that cheap data collection based on web traffic monitoring enables us to create a model that fairly correlates traffic density with air pollution. Building multiple models according to the hourly occurrence of the pollution peaks seems to increase the accuracy of the estimation, especially in the morning hours. The second result shows that adding meteorological factors allows for a significant improvement of the prediction of PM2.5 concentrations. Nevertheless, the last finding demonstrates that the best predictive model should be based on a hybrid source of data that includes trace gases. Since the sensors to monitor such gases are costly, the last part of the chapter gives some recommendations to get an accurate prediction from models that consider no more than two trace gases.

**Keywords:** urban air pollution prediction, heterogeneous data sources, hybrid models, low-cost approach, real-time traffic monitoring, meteorological and chemical features
