API enabled AQI Prediction using Federated Learning Presented by, Anurag Singh B200058CS Under the Supervision of Dr. Pankaj Kesarwani National Institute of Technology Sikkim Department of Computer Science Anurag Singh National Institute of Technology Sikkim 12 October 2023 Content • • • • • • • • • Introduction Literature Survey Research Gap Motivation & Objective Problem Statement Pre-requisite Current Work Future Work References 12 October 2023 Introduction 12 October 2023 Importance of AQI prediction 1.Public Health: Accurate AQI prediction helps protect public health by providing early warnings about deteriorating air quality, allowing individuals to take necessary precautions. 2.Environmental Protection: AQI prediction supports environmental conservation by identifying sources of pollution, enabling regulatory agencies to implement measures to reduce emissions. 3.Policy Development: Reliable AQI forecasts inform the development of air quality regulations and policies, helping governments make informed decisions to mitigate pollution. 4.Resource Allocation: Predicting AQI aids in allocating resources efficiently, as it enables authorities to focus efforts and resources on areas with the greatest need for pollution control. 5.Quality of Life: Access to reliable AQI forecasts enhances the overall quality of life by enabling individuals and communities to plan outdoor activities and reduce exposure to harmful pollutants. Introduction 12 October 2023 Federated Learning 1.Privacy-Preserving Collaboration: Federated learning allows various air quality monitoring stations, both public and private, to collaborate without sharing sensitive data. Each station trains a local model on its data, and only model updates are shared, preserving privacy. 2.Localized Accuracy: By integrating data from multiple sources, federated learning provides more comprehensive insights into air quality conditions in different locations. This approach helps in building location-specific models that capture local variations effectively. 3.Efficiency and Resource Utilization: AQI prediction models benefit from federated learning's ability to optimize model parameters across diverse datasets. This method enhances prediction accuracy, reduces computational costs, and efficiently utilizes data resources from various monitoring stations. 12 October 2023 Literature Survey S.no Author Title Specification Limitation 1 J. Ma, et al [1] Air quality prediction at new stations using spatially transferred bi-directional long short-term memory network Transfer Learning in Stacked BLSTM boosts air pollution prediction accuracy for new stations with limited data by transferring existing knowledge. Performance relies on ample pre-training data from existing stations, which can be limited in some regions. 2 S.A Janabi et al. [2] A new method for prediction of air pollution based on intelligent computation In SAQPM, PSO optimizes network parameters, enhancing accuracy via architecture fine-tuning in combination with RNN (LSTM). The blend of DSN-PS and DLSTM elevates computational complexity, causing longer execution times, impacting real-time predictions. A deep learning and image-based model for air quality estimation AQC-Net, a deep learning model using scene images, improves air quality estimation for a holistic understanding. AQC-Net's accuracy improvement involves increased model complexity, demanding more computational resources for training and deployment. LSTM for PM10 and SO2 prediction outperforms MLP and RNN models. Focused on PM10 and SO2; didn't explore meteorological influences on deep learning models for broader assessment. 3 4 Q Zhang et al. [3] B Das et al. [4] Prediction of air pollutants for air quality using deep learning methods in a metropolitan city 12 October 2023 Literature Survey S.no Author Title Specification Limitation 5 M Krishan, et al. [5] Air quality modelling using long short-term memory (LSTM) over NCT-Delhi, India LSTM models considering meteorological conditions, traffic data, and pollutant levels prove efficient and effective for air pollutant prediction, informing urban air quality research and policy. Study focuses on Delhi's pollutants, limiting generalizability; efficient LSTM models may need substantial resources, posing constraints. 6 P S Mahesh, et al. [6] A Survey Paper on Air Pollution using IoT. Enables real-time monitoring, cost-effectiveness, wide applicability, and mobile potential. Dependent on data thresholds, gas detection limitations, and internet connectivity. K Kumar, et al. [7] Air pollution prediction with machine learning: a case study of Indian cities Indian city air quality prediction with ML, addressing imbalance, XGBoost excels. Overlooks deep learning, limiting advanced modeling exploration. 7 Research Gap 12 October 2023 • Research gaps include the need to explore advanced deep learning methods in air quality prediction, particularly in developing countries like India, and to address the limitations of traditional methods. • The potential benefits of integrating real-time data and IoT technologies for more robust air quality monitoring and prediction have not been explored. • While IoT-based air pollution monitoring systems show promise, there is limited discussion about potential challenges, such as data security and privacy. • Adapting LSTM models for air pollutant prediction to different locations and pollutants with varying characteristics and sources is not addressed. • The research could benefit from exploring hybrid approaches that combine deep learning with traditional methods to enhance prediction accuracy. Motivation & Objective 12 October 2023 • Enhanced Model Training: The motivation behind employing Federated Learning in air quality prediction is to facilitate collaborative, location-specific model training, addressing the challenge of varying pollutant characteristics and sources across regions which are not done in previous works. • Privacy-Preserving Data Sharing: To ensure data privacy and security while enabling effective data sharing for air quality monitoring, thereby mitigating concerns related to IoT technology and data security. • Efficient Model Updates: Motivation stems from the need to efficiently update air quality prediction models with real-time data, making it possible to harness the benefits of real-time data integration. Motivation & Objective 12 October 2023 • Robust and Generalizable Models: The objective is to develop robust and generalizable air quality prediction models by aggregating knowledge from diverse locations through Federated Learning. • Preservation of Data Privacy: The primary objective is to preserve the privacy of air quality data while enabling collaborative model training, ensuring that sensitive information remains secure. • Hybrid Approaches: The objective is to explore the synergies of combining deep learning with traditional methods in air quality prediction while leveraging Federated Learning to enhance accuracy, adaptability, and data privacy. Problem Statement 12 October 2023 The study focuses on advancing Air Quality Index (AQI) prediction by harnessing Federated Learning and Application Programming Interfaces (APIs). Challenges in air quality forecasting, such as location-specific adaptability, privacy concerns, and realtime data integration, necessitate innovative solutions. The research employs Federated Learning, a privacy-preserving method, to create an API-enabled AQI prediction system. This system enhances accuracy, maintains data privacy, and explores hybrid modeling, addressing the complexities of air quality prediction, especially in locations with varying pollutant characteristics and sources. The work contributes to more effective air quality monitoring and management. Pre-requisite 12 October 2023 In addition to the previous attributes mentioned, the following factors may also be considered for AQI prediction: • Historical AQI data: This can help the model to learn the patterns of air pollution over time. • Chemical data: This can include data on the chemical composition of the air pollution. • Human activity data: This can include data on traffic patterns, industrial activity, and energy consumption. Pre-requisite 12 October 2023 The following attributes are required for AQI prediction: • Air pollutant concentrations: This includes the concentrations of particulate matter (PM2.5 and PM10), ozone (O3), nitrogen dioxide (NO2), sulfur dioxide (SO2), and carbon monoxide (CO). • Meteorological data: This includes temperature, humidity, wind speed and direction, and precipitation. • Topography: This includes the elevation and land cover of the area. • Emission sources: This includes the location and type of emission sources, such as vehicles, industry, and power plants. Current Work Field Descriptors of AQI attributes fetched using APIs: • • • • • • • • • • • • • lat: Latitude (Degrees). lon: Longitude (Degrees). timezone: Local IANA Timezone. city_name: City name. country_code: Country abbreviation. state_code: State abbreviation/code. aqi: Air Quality Index [US - EPA standard 0 - +500] o3: Concentration of surface O3 (µg/m³) so2: Concentration of surface SO2 (µg/m³) no2: Concentration of surface NO2 (µg/m³) co: Concentration of carbon monoxide (µg/m³) pm25: Concentration of particulate matter < 2.5 microns (µg/m³) pm10: Concentration of particulate matter < 10 microns (µg/m³) 12 October 2023 Current Work Sample attributes of AQI fetched using API: 12 October 2023 Current Work Field Descriptors of Meteorological attributes fetched using APIs: • • • • • • • • • • Latitude: Latitude in degrees. Longitude: Longitude in degrees. Time: Timestamp or observation moment. Temperature at 2m: Air temperature at 2 meters. Relative Humidity at 2m: Moisture percentage at 2 meters. Dew Point at 2m: Saturation temperature at 2 meters. Apparent Temperature: "Feels-like" temperature. Precipitation: Total moisture (rain, snow). Rain: Liquid precipitation amount. Snowfall: Snow depth or amount. 12 October 2023 Current Work Field Descriptors of Meteorological attributes fetched using APIs: • • • • • • • • • • Pressure at MSL: Sea-level atmospheric pressure. Surface Pressure: Surface atmospheric pressure. Cloud Cover: Sky coverage percentage. Cloud Cover (Low, Mid, High): Coverage at different levels. ET0 (FAO Evapotranspiration): Evaporation/transpiration rate. Wind Speed at 10m: 10m wind speed. Wind Speed at 100m: 100m wind speed. Wind Direction at 10m: 10m wind direction. Wind Direction at 100m: 100m wind direction. Wind Gusts at 10m: Sudden wind speed increases. 12 October 2023 Current Work Sample attributes of Meteorological data fetched using API: 12 October 2023 Future Work • • • • • • • • Federated Learning Infrastructure Collaborative Model Design Communication and Aggregation Location-Specific Adaptability Real-Time Data Integration Model Evaluation API Integration Testing and Validation 12 October 2023 References 12 October 2023 [1] Ma, Jun, Zheng Li, Jack CP Cheng, Yuexiong Ding, Changqing Lin, and Zherui Xu. "Air quality prediction at new stations using spatially transferred bi-directional long short-term memory network." Science of The Total Environment 705 (2020): 135771. [2] Al-Janabi, Samaher, Mustafa Mohammad, and Ali Al-Sultan. "A new method for prediction of air pollution based on intelligent computation." Soft Computing 24, no. 1 (2020): 661-680. [3] Zhang, Qiang, Fengchen Fu, and Ran Tian. "A deep learning and image-based model for air quality estimation." Science of The Total Environment 724 (2020): 138178. [4] Das, Bihter, Ömer Osman Dursun, and Suat Toraman. "Prediction of air pollutants for air quality using deep learning methods in a metropolitan city." Urban Climate 46 (2022): 101291. [5] Krishan, Mrigank, Srinidhi Jha, Jew Das, Avantika Singh, Manish Kumar Goyal, and Chandrra Sekar. "Air quality modelling using long short-term memory (LSTM) over NCT-Delhi, India." Air Quality, Atmosphere & Health 12 (2019): 899-908. [6] Mahesh, Pattar Sunil, Patil Bhushan Rajendra, Bodke Akshay Dnyaneshwar, and Mr Ulhās V. Patil. "A survey paper on air pollution monitoring using IOT." (2018). [7] Kumar, K., and B. P. Pande. "Air pollution prediction with machine learning: a case study of Indian cities." International Journal of Environmental Science and Technology 20, no. 5 (2023): 5333-5348. Thank You