《2020年通过数据集成和人工智能绘制贫困地图:亚太地区关键指标补充 - 亚洲开发银行(英文版)(45页).pdf》由会员分享,可在线阅读,更多相关《2020年通过数据集成和人工智能绘制贫困地图:亚太地区关键指标补充 - 亚洲开发银行(英文版)(45页).pdf(55页珍藏版)》请在三个皮匠报告上搜索。
1、ASIAN DEVELOPMENT BANK A Special Supplement of the Key Indicators for Asia and the Pacific 2020 MAPPING POVERTY THROUGH DATA INTEGRATION AND ARTIFICIAL INTELLIGENCE SEPTEMBER 2020 ASIAN DEVELOPMENT BANK A Special Supplement of the Key Indicators for Asia and the Pacific 2020 MAPPING POVERTY THROUGH
2、DATA INTEGRATION AND ARTIFICIAL INTELLIGENCE SEPTEMBER 2020 Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) 2020 Asian Development Bank 6 ADB Avenue, Mandaluyong City, 1550 Metro Manila, Philippines Tel +63 2 8632 4444; Fax +63 2 8636 2444 www.adb.org Some rights reserved. Published in
3、2020. ISBN 978-92-9262-313-5 (print); 978-92-9262-314-2 electronic); 978-92-9262-315-9 (ebook) Publication Stock No. FLS200215-3 DOI: http:/dx.doi.org/10.22617/FLS200215-3 The views expressed in this publication are those of the authors and do not necessarily reflect the views and policies ofthe Asi
4、an Development Bank (ADB) or its Board of Governors or the governments they represent. ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The mention of specific companies or products of manufacturers does not
5、imply that they are endorsed or recommended by ADB in preference to others of a similar nature that are not mentioned. By making any designation of or reference to a particular territory or geographic area, or by using the term “country” inthis document, ADB does not intend to make any judgments as
6、to the legal or other status of any territory or area. This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) https:/creativecommons.org/licenses/by/3.0/igo/. By using the content of this publication, you agree to be bound bytheterms of this license. For attrib
7、ution, translations, adaptations, and permissions, please read the provisions andterms of use at https:/www.adb.org/terms-use#openaccess. This CC license does not apply to non-ADB copyright materials in this publication. If the material is attributed toanother source, please contact the copyright ow
8、ner or publisher of that source for permission to reproduce it. ADB cannot be held liable for any claims that arise as a result of your use of the material. Please contact pubsmarketingadb.org if you have questions or comments with respect to content, or if you wish toobtain copyright permission for
9、 your intended use that does not fall within these terms, or for permission to use theADB logo. Corrigenda to ADB publications may be found at http:/www.adb.org/publications/corrigenda. Notes: In this publication, “$” refers to United States dollars, unless otherwise stated. Cover design by Rhommell
10、 Rico. iii Contents Foreword . v Highlights .vii Introduction .1 Estimating Poverty Using Conventional Data Sources .3 Using Big Data to Enhance Development Statistics .5 Predicting Poverty Using Geospatial Data .6 Using Neural Networks to Develop an Algorithm .10 Understanding Convolutional Neural
11、Networks .12 Outlining Data Requirements for the Feasibility Study .13 Daytime satellite imagery .13 Data on night lights.15 Poverty statistics .17 Applying a Convolutional Neural Network to Poverty Prediction .18 Extracting Features from the Convolutional Neural Network .20 Predicting Poverty from
12、Features Using Ridge Regression Models .21 Outlining the Key Findings of the Feasibility Study .22 Preparing National Statistics Offices for the Use of Big Data .29 Access to nontraditional data .33 Technological requirements .33 Capabilities .34 Data privacy .34 Ecosystem .34 Institutional strength
13、ening .35 Summary and Conclusion .35 References .37 Appendixes Appendix 1: A Primer on Deep Learning Concepts .40 Appendix 2: Sources of Satellite Imagery .43 ivA Special Supplement of the Key Indicators for Asia and the Pacific 2020 Tables, Figures, and Boxes Tables 1 Select Countries in Asia and t
14、he Pacific that Use the World Banks Poverty Mapping Methodology .5 2 Summary of Poverty Prediction Results using Confusion Matrixes, Thailand .22 3 Summary of Poverty Prediction Results using Confusion Matrixes, Philippines .22 Appendix Tables A1.1 Sample of a Typical Confusion Matrix .41 A1.2 Compa
15、rison of Popular Deep-Learning Software Communities .43 A2.1 Uses of Landsat 8 Spectral Bands .44 Figures 1 Illustration of a Computer Vision Task .8 2 Illustration of a Neural Network .11 3 Neural Network Filters to Detect Vertical and Horizontal Lines .12 4 Road Map of Methodology for Predicting P
16、overty Using Satellite Imagery .14 5 Pansharpening Images to Improve Their Resolution .15 6 Night Light Image Tiles for the Philippines and Thailand .16 7 Distribution of Poverty Rates in the Philippines and Thailand .18 8 Image Color Bands within a Georeferenced Image File .18 9 Low-Quality Satelli
17、te Images Isolated from Training .19 10 Examples of Features Extracted from the Convolutional Neural Network .20 11 Extracting a Convolutional Neural Networks Output Layer .21 12 Scatter Plot of Published and Predicted Poverty Rates .26 13 Maps of Published and Predicted Poverty Rates .30 14 Calibra
18、ted Machine-Learning Poverty Predictions .32 Appendix Figures A1.1 Illustration of a Sample Neural Network .41 A1.2 Measuring Cross Entropy Loss against Predicted Probability .41 Boxes 1 Integration of Big Data into a Small Area Estimation Framework .6 2 Two Broad Methods of Predicting Poverty Using
19、 Geospatial Data .7 3 Other Development Applications of Night Light Data .9 4 Compiling Grid-Level Estimates of Poverty Head Count .23 5 Does the Algorithms Prediction Accuracy Improve when the Indicator has more Variability? .27 v Foreword Can nontraditional data sources, such as satellite imagery,
20、 serve as a useful supplementary data source in measuring the Sustainable Development Goals (SDGs)? Economies worldwide are racing to meet the SDGs by 2030. Pledging to leave no one behind, and to first assist those most in need, the 17 SDGs are monitored through a global framework consisting of 155
21、 targets and 231 unique indicators. The majority of these indicators are compiled by national statistical systems through representative surveys, censuses, and administrative records. For instance, the indicators for SDG 1: Eradication of Poverty conventionally come from surveys on household income
22、and expenditure, or on living standards. Often, such surveys have sample sizes that are large enough to provide nationally representative estimates. These sample sizes generally also provide estimates that fall within tolerable levels of reliability when further disaggregating poverty estimates by s
23、ignificant or established intranational domains, e.g., by states, provinces, or regions. However, survey sample sizes are typically not large enough to provide reliable estimates at more granular levels, such as municipalities and villages, and therefore may not be able to assist policymakers in eff
24、iciently targeting population segments that have the greatest need for poverty reduction programs. Increasing survey sample sizes to produce reliable estimates at granular levels is the ideal option, but it is often not practical. Achieving such increases requires significant additional resources, w
25、hich are not readily available to national statistics offices (NSOs) or the organizations that conduct national surveys. As an alternative, some countries adopt small area estimation methods, whereby survey data are complemented with auxiliary data from census or administrative records. These auxili
26、ary data sources are very useful because they facilitate further disaggregation of poverty statistics to provide reliable estimates at more granular levels than those for which surveys might be originally designed. However, since census and administrative data are frequently not available or readily
27、 accessible, there are studies that explore the use of alternative sources of auxiliary data because such data are not prone to sampling errors. In 2017, the Asian Development Bank (ADB) designed a knowledge initiative called Data for Development, which aims to strengthen the capacity of NSOs in the
28、 Asia and Pacific region to meet the increasing data demands for effective policymaking and for monitoring development goals and targets. One component of the initiative focuses on subnational disaggregation of SDG indicators, particularly poverty statistics. This component draws inspiration from st
29、udies that use high resolution satellite imagery, geospatial data, and powerful machine-learning algorithms to complement traditional data sources and conventional survey methods. This approach can be used to estimate the magnitude of poverty in specific areas in the world, and the resulting data ca
30、n aid governments and development organizations in distributing funds more efficiently as well as helping policymakers design more effective and targeted poverty reduction strategies. Statisticians from ADBs Statistics and Data Innovation Unit within the Economic Research and Regional Cooperation De
31、partment worked with the Philippine Statistics Authority, the National Statistical Office of Thailand, and the World Data Lab to examine the feasibility of poverty mapping using satellite imagery and associated geospatial data. viA Special Supplement of the Key Indicators for Asia and the Pacific 20
32、20 This supplement to Key Indicators for Asia and the Pacific 2020 documents the initial results of the feasibility study, which aimed to explore alternative data collection channels by combining traditional methods with innovative sources that might enhance the granularity, cost effectiveness, and compilation of high-quality poverty statistics. The publication team was led by Arturo Martinez, Jr, under the over all direction of Elaine Tan. Arturo Martinez, Jr., Mildred Addaw