Analysis and Prediction of Air Quality Data based on Modularized SVM Classifier

• Main Authors: Chien-Fu Huang, 黃建富
• Other Authors: 陳彥霖
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/sxnzrq

碩士 === 國立臺北科技大學 === 資訊工程系 === 106 === The problem of air pollution has gradually been paid more attention in recent years. Air pollution sources such as PM2.5, formaldehyde, carbon monoxide or excessive carbon dioxide may cause asthma, lung cancer or cardiovascular diseases after prolonged exposure. Most of the current standard values of air quality analysis, prediction are mainly focused in outdoor large area or a aingle city. The air quality indicators used in this paper will be different from outdoor conditions. Therefore, this thesis proposea a machine learning technique and designa a framework for air quality analysis and prediction, through the air pollution sensor combined with the LoRa transmission module and the MQTT transmission protocol to collect data for long-term indoor air pollution values. Since the collection of data for a long time must produce huge data, this study adopted such a huge amount of data to estimate the standard value of air quality in the region. It is expected that the indoor space will have more air pollution sources at certain time spans. The system can automatically turn on specific air purification equipment at this time, or use the IoT to automatically turn the air purification unit on or off. The previous machine learning frameworks are used for specific systems. This thesis aims to modularize the air quality analysis and prediction machine learning framework. This paper uses SVR to predict and analyze air quality. The paper will complete a modular system based on DAG-SVM multi-classifier, which provides modules for image or data input for classification. At the same time, in order to apply to this research, the DAG-SVM multi-classifier will be modified to perform air quality anlysisa, and uses the SVR method to predict the air quality for the modular system. This study use SVM multi-classifier to classify whether the current air quality is abnormal. The air quality analysis through SVR can know which value the air pollution is expected to fall at this time point in the next cycle, and the error rate is less than 10%. The thesis modularizes the SVM multi-classifier and proposes a system framework for friendly construction of machine learning system. Users can easily replace the SVM module to train and detect various SVM classifiers.