Scholarlink

<p>Due to rapid urbanization, pollution levels have tremendously increased in recent times, degrading the quality of life in cities. If pollution levels are hazardous, they can seriously affect human health. Online machine learning algorithms are available that can be deployed for real-time air quality monitoring using real-time streaming data. These online algorithms are used to train models that detect the presence of anomalies in the incoming data stream, thereby improving reliability and accuracy. In this research, nine popular unsupervised online anomaly detection models are used to analyze the Air Quality Index (AQI) of Delhi, the capital of India. Gaseous pollutants are monitored in real time, for the presence of anomalies, at 20 base stations corresponding to the most polluted sites in the city in the duration of 1st May 2022 to 30th April 2023. The models include Half Space Trees (HST), Quantile Filter One-Class Support Vector Machine (QF_OCSVM), Quantile Filter Half Space Trees (QF_HST), Threshold Filter Half Space Trees (TF_HST), Histogram Based Outlier Score (HBOS), Kitsune’s Core Algorithm (KitNet), Robust Random Cut Forest (RRCF), Isolation Forest Anomaly Streaming Data (IForestASD), and K-Nearest Neighbor Conformal Anomaly Detection (KNNCAD). The performance is evaluated in terms of four metrics: Accuracy, ROCAUC, Recall, and Geometric Mean. We analyze the performance of each model with timestamps versus anomaly score plots. The QF_OCSVM, QF_HST, and HBOS models achieved the highest scores across four performance metrics, while the remaining models show stable performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Real-time Air Pollution Monitoring in Delhi Using Online Anomaly Detection Algorithms for Streaming Data

  • 期刊论文

  • 发表日期：

    2025年7月22日

查看全文
  • Santosh Kumar Ray,
  • Seba Susan

摘要

Due to rapid urbanization, pollution levels have tremendously increased in recent times, degrading the quality of life in cities. If pollution levels are hazardous, they can seriously affect human health. Online machine learning algorithms are available that can be deployed for real-time air quality monitoring using real-time streaming data. These online algorithms are used to train models that detect the presence of anomalies in the incoming data stream, thereby improving reliability and accuracy. In this research, nine popular unsupervised online anomaly detection models are used to analyze the Air Quality Index (AQI) of Delhi, the capital of India. Gaseous pollutants are monitored in real time, for the presence of anomalies, at 20 base stations corresponding to the most polluted sites in the city in the duration of 1st May 2022 to 30th April 2023. The models include Half Space Trees (HST), Quantile Filter One-Class Support Vector Machine (QF_OCSVM), Quantile Filter Half Space Trees (QF_HST), Threshold Filter Half Space Trees (TF_HST), Histogram Based Outlier Score (HBOS), Kitsune’s Core Algorithm (KitNet), Robust Random Cut Forest (RRCF), Isolation Forest Anomaly Streaming Data (IForestASD), and K-Nearest Neighbor Conformal Anomaly Detection (KNNCAD). The performance is evaluated in terms of four metrics: Accuracy, ROCAUC, Recall, and Geometric Mean. We analyze the performance of each model with timestamps versus anomaly score plots. The QF_OCSVM, QF_HST, and HBOS models achieved the highest scores across four performance metrics, while the remaining models show stable performance.