An enhanced air quality prediction of low-cost air quality sensors dataset in Rouen (France) using artificial intelligence with principal component analysis as feature learning for Sustainable Development Goals (SDGs)

¹ Graduate Institute of Automation and Control, National Taiwan University of Science and Technology, Taipei, Taiwan.
² Department of Physics, Universitas Negeri Surabaya, Indonesia

^* Corresponding author: chihtayen@gmail.com

Abstract

Air quality has been a recurring topic of controversy during the last five years. One part of supporting SDG 13 (climate action) is research into the impact of air quality and how to address it. The majority of artificial intelligence prediction research makes use of datasets that have been confirmed for AQI levels. However, few researchers manually process and analyze the raw sensor measurement record. The primary goal of this work is to use feature extraction principal component analysis to increase the accuracy of predictions made using actual sensor data from Rouen, France. Data preparation and feature extraction are critical for the model to avoid overfitting. The original sensor data contains nine observations from October 2021 to March 2022. Too many features might also lead to overfitting. Therefore, suitable data preparation is required. Several machine learning and Deep learning are examples of models used to demonstrate the effect of feature extraction. This investigation produced remarkable results, with the performance of the five models increasing by around 96.72% to 99.35% when compared to similar experiments using data confirmed by the observation agency.