Volkova Nataliya, Senior project manager of Analytics Department at the IT Department of Moscow Government prepared an article on the applications of predictive analytics and decision support systems for Smart City, published in the Artificial Intelligence Almanac #5 “Predictive analytics and decision support systems” in August 2020.
The idea of smart cities is not new - it emerged over forty years ago in California. In 1974, the Los Angeles Community Analysis Bureau released the “State of the City: Cluster Analysis of Los Angeles” report, which contained proposals for the creation of a city-wide information and analytical system for metropolitan management “on a scientific basis”. And back in the late 60s, the Bureau undertook the first and rather successful attempt to apply mathematical and software models to solve the urgent problems of the metropolis: combating crime, reducing road accidents, helping low-income population, decommissioning dilapidated housing - in total, 16 problem areas were identified.
However, at the global level the topic of smart sustainable cities took root much later, when in 2012 the United Nations Economic Commission for Europe included it in the work program. Two years later, the United Nations project “United Smart Cities” (USC) was launched to support sustainable urban development with a focus on providing access to innovation technologies.
During the implementation of the USC project, numerous consultations with technology companies, the academic community and government organizations were held, over 300 industry experts were interviewed, 116 definitions of a “smart city” were analyzed. As a result, the UN and ITU (specialized agency of the United Nations responsible for all matters related to information and communication technologies.) prepared several guiding documents and created a single definition: “A smart sustainable city is an innovative city that uses information and communication technologies (ICTs) and other means to improve quality of life, efficiency of urban operation and services, and competitiveness, while ensuring that it meets the needs of present and future generations with respect to economic, social, cultural and environmental aspects”.
Big data, open data and anonymization infrastructure, technologies for ensuring data security, as well as predictive analytics and decision support systems are among the most significant technologies, according to the UN-ITU reports.
In 2019 the USC project was integrated into the United for Smart Sustainable Cities (U4SSC) initiative, which is aimed at supporting projects using advanced data processing and analysis technologies to achieve UN Sustainable Development Goals (SDGs).
Numerous rankings, such as Smart City Index, Cities in Motion, Innovation Cities Index and Sustainable Cities Index, do not provide a clear-cut answer to this question. Quite often the first places are taken by the cities in which the organizations that compile the ranking are located.
An examination of the rankings’ methodology shows that the availability of big urban data platforms and analytical systems is usually not a selection criterion. At the same time, data-driven governance helps cities to be in the top of the rankings. The use of analysis and forecasting tools is taken into account indirectly, through the impact on environmental, social, transport and other aspects of the urban environment.
Let us consider some of the most interesting examples of how old-timers and newcomers are tackling urban challenges using predictive analytics and machine learning. The UN carries out the monitoring of such projects as part of the U4SSC “City Science Application Framework”.
Certain counties in California experience high levels of air pollution from motor vehicle traffic, industrial emissions, forest fires, and other causes. Air quality control is regulated at the regional level across 35 regional pollution districts. The Bay Area Air Quality Management District (BAAQMD) operates in the San Francisco Bay Area, and South Coast Air Quality Management District (SCAQMD) operates in several Southern California counties (Los Angeles, Orange, Riverside and San Bernardino). Both districts use Envirosuite digital platform.
Before the introduction of the platform, data from sensors was being processed (which took quite a long time) before being transferred to responsible services. This approach has hampered rapid response to air quality incidents. In some cases, by the time the causes and sources of emissions were found, time had already been lost.
Envirosuite was deployed to solve this problem. The system in real time creates a baseline scenario for the spread of harmful substances in the atmosphere based on a set of data on air quality, wind speed and direction, industrial emissions and takes into account the weather forecast. The dataset is formed based on the readings of the sensors of air quality and weather stations connected to Envirosuite via wireless communication channels. The platform helps to identify potential sources of air pollution within seconds after an incident occurs.
The system uses algorithms (CALPUFF - California Puff Model, CALMET - Computer Aided Learning in Meteorology, WRF - Weather Research and Forecasting) based on mathematical models of the processes of dispersion of harmful substances in the atmosphere, and machine learning methods. Envirosuite runs on Amazon Web Services cloud infrastructure.
With the help of the system, California air quality control authorities are able to quickly respond to incidents and complaints from residents, develop effective plans to improve the environmental situation and create a more sustainable regional development model. Improving air quality will have a positive impact on the health of citizens and increase life expectancy. The project is currently being implemented, so the final results will be obtained in the coming years.
Dubai aims to be the happiest city in the world. And these are not utopian dreams, but the specific goal of the strategy of building a “smart city”. It is worth mentioning that the UAE is the only country in the world that has a Ministry of Happiness. At the same time, there is also an understanding that happiness is a subjective concept, and increasing the level of happiness is a task that is difficult to convert into traditional measures of state planning.
Non-standard goals require non-standard approaches. And the authorities of Dubai decided to measure the satisfaction level of the population and business in all major aspects of urban life, including education, healthcare, transport, economy, energy, and environment.
The surveys were conducted using the “Happiness Meter” - a mobile application for smartphones, in which users can rate the services of the public or private sector on a simple scale: happy/unhappy/neutral. In addition, residents and visitors of Dubai were encouraged to provide feedback, ideas for improvements and critical comments. Happiness Meter was also installed on information kiosks and added to city web resources.
Over the years, a lot of data concerning people’s daily lives has been collected: commuting to work, obtaining medical prescriptions, enrolling in school, buying and renting real estate, starting a business. To understand the correlation of estimates with the parameters of the service provided, the results were processed using statistical analysis and machine learning methods
Results of measuring the level of happiness in 2015-2018
From its launch in 2015 to the end of 2018, the Happiness Meter has collected over 22 million ratings of happiness and over 650 thousand reviews, which is quite a lot for an emirate with a population of 4 million people.
Compared to social media surveys and sentiment analysis, which are also used in urban analytics, the Happiness Meter platform and mobile app have become a unique tool for the Dubai authorities to provide information from various perspectives. The analytics platform helps the authorities and the private sector pursue a policy of continuous improvement of the quality of services, focus on problem areas, improve social services for the population, involve residents in city management, and increase the transparency of government activities. It is noted that the use of this tool contributes to the formation of a healthy competitive environment (companies compete for high ratings from consumers) and the faster dissemination of the “best practices”.
Rio de Janeiro: CrimeRadar for crime forecasting Shortly before the Summer Olympic Games in Rio de Janeiro, the City Hall launched the CrimeRadar digital platform and app. Local residents and tourists were able to see the forecast of the crime situation by district on the heat map of the city.The ability to forecast the place and time of crimes sounds fantastic and resembles the plot of Steven Spielberg’s blockbuster “Minority Report”. Meanwhile, the governments of North America, Western Europe and some Asian countries are actively developing this direction of predictive analytics.High crime rates have always been a serious problem in Rio. Law enforcement agencies are testing new methods for identifying hotbeds of crime, increasing the efficiency of resource allocation and creating optimal routes for police patrols. CrimeRadar is a digital situational analysis and crime forecasting platform that uses machine learning and Bayesian statistical methods. The model was trained on a five-year retrospective police data provided by the Brazil’s Institute for Public Security.The project was inspired by the Igarapé Institute (a think tank in Rio devoted to evidence-based policy on security and development challenges), which was engaged in the selection of platform and application developers, preparing datasets for ML-models training and models testing.
CrimeRadar allows users to visualize the crime situation in various areas of Rio de Janeiro. The probability of crime occurrence is displayed in sectors of 250x250 meters on a ten-point scale, where 1 is the safest zone (green) and 10 is the most dangerous (red). The forecast for the future is calculated by days of the week and time of day. Unlike similar systems in other countries, the heat map of Rio’s crime rate is available not only to law enforcement agencies, but also to the residents of the city.
One of the features of the project was the inclusion of liability provisions in the platform software license (including compensation to citizens for damage caused by an incorrect forecast). The license contained a reference to the code of ethics for the use of AI prepared by FAT/ML (Fairness, Accountability, and Transparency in Machine Learning), a community of researchers concerned with fairness, accountability and transparency in machine learning. However, a disclaimer was placed on the crime map with a recommendation not to rely entirely on predictive algorithms when making decisions on personal security issues. The disclaimer indicates that the model is not currently being updated and its accuracy is decreasing.
According to the Igarapé Institute, the implementation of the CrimeRadar helps to curb the growth of crime, improve the efficiency of police work and improve the tourist image of Rio de Janeiro. At the same time, there are no statistical studies on the results of using this model, and randomized controlled trials of similar crime forecasting models provide mixed results. Critics of the system also point out that information about “green” zones gives residents a false sense of security and can be used by criminals to commit crimes.
1. Medical decision support in Unified Radiological Information Service (ERIS) In January 2020, at the XVIII Healthy Moscow Assembly, the City Health Department announced the start of an experiment to introduce computer vision for the analysis of medical images. The purpose of the experiment is to study the possibility of using medical decision support methods with innovative technologies in the Moscow healthcare system. The technological platform of the experiment is the Unified Radiological Information Service (ERIS), which contains datasets for training models for recognizing lung cancer, breast cancer, tuberculosis and other diseases. Developers have access to a database of images for various types of medical imaging. Participants are registered on the official portal of the experiment - mosmed.ai. The documentation and regulations for the competition are also available there. Clinical trials of computer vision systems are carried out in accordance with the guidelines developed by the Scientific and Practical Clinical Center of Diagnostics and Telemedicine Technologies of the Moscow Department of Health. Experiment participants receive grants, the amount of which depends on the number of analyzed medical images.
The portal of the experiment - mosmed.ai
During the experiment, Moscow faced the COVID-19 pandemic. The primary task was to create a service that automatically detects signs of coronavirus pneumonia in CT scans. The team of the Scientific and Practical Clinical Center of Diagnostics and Telemedicine Technologies led by Sergey Morozov formed and published a unique dataset “MosMedData: the results of chest CT scans with COVID-19 related findings”, which became the largest in the world. Currently, the dataset contains 1100 scans that have been performed in more than 80 medical institutions in Moscow connected to the ERIS. Studies are divided into 5 categories depending on the volume of pathological changes in the lungs.
Presentation to the report of prof. S.P. Morozov at the round table “Artificial Intelligence in Health Care ”
The results of the experiment are quite positive. Since the beginning of the epidemic, more than 200 thousand lung scans have been performed at Moscow’s computed tomography centers, and most of them have been analyzed using machine learning technologies.
The service based on artificial intelligence marks areas of possible lung damage in CT scans and forms a preliminary conclusion on the likelihood of pathology, which allows to reduce the time of the examination, increase the productivity of medical experts, reduce the amount of routine actions and improve the quality of diagnostics. But the final diagnosis is made by the doctor.
Traditionally, recommendation systems in smart cities are tools in state information systems aimed at supporting decision-making by civil officials. Meanwhile, urban recommendation systems are needed primarily by residents, in addition to personal assistants on their smartphones.
Since 2016 a virtual operator has been working in the Moscow citywide contact center - a speech recognition and synthesis system based on partner solutions. The operator provides recommendations on 30 topics: payment of utilities, readiness of documents, location of the evacuated vehicle and the procedure for its return, the address of the nearest center “My Documents” and others. In 2019, the robot processed 8.5 million calls. The training knowledge base of the speech recognition and synthesis system in the contact center currently contains more than 8.5 thousand articles.
The virtual operator reduced the load on the contact center operators by undertaking the responses to typical requests from citizens, and reduced the waiting time on the line for massive seasonal calls. The goals and tasks for the future are to teach the robot to conduct a full-fledged dialogue, the ability to understand the emotional state of the person and adjust to it, and predict the questions of the residents.
In 2019, at the “Crystal Headset” competition, Moscow virtual operator won the “Best Practice of Customer Service without Operator Involvement” and became a prize winner in the “Best Use of Automation, Robotics and AI” category.
Moscow is also testing systems that use predictive analytics in education. The ambitious overall goal of these projects is to improve educational performance of schoolchildren and increase the likelihood of university enrollment through a personalized recommendation service for additional training.
A student's “digital footprint” is a set of structured and unstructured data, first of all it includes grades in various subjects, attendance at lessons and extra classes, results of examinations, choice of clubs, participation in Olympiads and winnings, school rating - more than 100 indicators in total. Models are trained on retrospective data from 800 educational institutions of Moscow, allowing to predict the results of final exams and automatically generate recommendations for improving student performance, including with the help of materials from the Moscow Electronic School. To help both students and teachers, tools to visualize the relative performance of each pupil on radar charts have been implemented. The mean absolute error (MAE) of the prediction of the Unified State Exam (USE) score of Moscow pupils in the Russian language was 4.40, in Mathematics - 7.64.
On the basis of the results of the OGE (Basic State Examination) and USE, the number of participants and winners of the Olympiads, interest in project activities in the subject of the teacher, the best teaching personnel are identified.
Future plans include
improving the accuracy of the models by enriching the datasets with additional
information about the psychological characteristics of the personality of
students based on the results of questionnaires, training with tutors, and
other factors outside the public education system. It is also planned to
develop tools for helping schoolchildren in vocational guidance, choosing a
profile school and university program. The studies of the correlation between
the USE score and the training manual are also underway. In the future, the
system will be able to provide guidance on the selection of training materials
for each of the subjects.
Traditionally, the order of work in particular houses is determined by regional capital repair programs based on the duration of operation of the house and its engineering systems (from the moment of commissioning or previous repairs) and the assessment of the technical condition of the house based on monitoring results. Time between overhaul and other parameters are established in technical and regulatory documents (Building code, territorial building codes, technical conditions, etc.). However, in some cases, there is a need to carry out repairs ahead of the prescribed time.
The application of machine learning methods is a new approach to this problem. The data set was formed on the basis of house passports with location, physical and technical characteristics, information on actually carried out repairs, and it is enriched with readings of sensors and requests from residents for repairs to the Unified Dispatch Center. The training sample includes data from 8.5 thousand houses, about 90 thousand contracts for repairs in more than 8 thousand houses, 6 million requests from residents of 30 thousand houses for 27 types of required repairs: elevators, plumbing, sewage, roofing, etc. The “Elevator breakdown prediction” (LightGBM) model used 20 indicators.
As a result of the pilot project, a rating of buildings recommended for inclusion in the capital repair program was formed (Gradient boosting (LightGBM, Catboost) was used for prediction. The prediction accuracy of the Elevators model according to ROC_AUC was 0.95, and the Roof model was 0.65.)
The Moscow Suppliers Portal is a digital platform for automating small-scale procurement activities, a kind of online store. 194 thousand suppliers from 36 regions of Russia are registered on the resource, the catalogue contains more than half a million standard trade items.
Tracking purchases “in manual mode” (there are about 40 thousand of them every month) is a resource-intensive process, despite the advanced filtering function. Therefore, in 2017 a recommendation service was launched on the portal. It selects potentially interesting purchases for entrepreneurs and sends them a personalized mailing. The initial data for the model are published procurement plans, announced procurement, and the supplier’s history on the platform.
A similar service is implemented on the portal of the Moscow Innovation Cluster. Based on the semantic analysis of user requests on the i.Moscow website, the system recommends business partners to cluster members. Other recommendation services are also being tested, including the search for infrastructure, investors, support measures, technologies and results of scientific research.
The city is also testing other models of forecasting and decision-making support: tools to support planning of urban infrastructure facilities construction, prediction of the resource consumption in housing and public utilities, identification of crime hotspots and analysis of operational information, construction of a crime map. With the help of video surveillance technologies and video analytics, two thirds of all reported crimes are solved in Moscow. Analysis of citizens’ appeals, their classification and referral to specialized executive authorities are automated.
Human nature does not change - since ancient times, people have been striving to look into the future. But if in the past people turned to the Delphic oracles and the prophecies of Nostradamus, now they turn to datascientists and the methods of predictive analytics.
In the article “Predicting is the silliest thing to do”, the founder of Soviet cosmonautics, academician Boris Rauschenbach wrote that “Human life evolves under very complex laws, non-linear, as mathematicians put it, and we can only predict linearly; a direct continuation is useful for a couple of years, well, for 10 years, not more [...] Therefore, nothing can be predicted, even for a professional in the scientific field”.
At the same time, technology giants employ famous futurists, and Raymond Kurzweil’s work at Google is the most glaring example of this. Developers of predictive technologies in healthcare, epidemiology, social and financial spheres were invited to the UN Group on Digital Cooperation: Dr. Kira Radinsky, Sophie Soowon Eom and others.
Of course, rapid social and technological transformations in all spheres of life further narrow the horizons of forecasting. And at the same time, it creates new predictive methods and data that that allow to see the future. Modern smart cities have a large number of big data sources: intelligent transport systems, Internet of Things devices, state information systems and video surveillance platforms. In this context, data serves as a strategic asset of megacities, with the help of which they can solve federal and regional problems, improve the urban environment and contribute to the achievement of the UN global sustainable development goals.