Akademik Veri Yönetim Sistemi
International Conference on Recent Academic Studies ICRAS 2023, Konya, Türkiye, 2 - 04 Mayıs 2023, cilt.1, sa.29802075, ss.85, (Özet Bildiri)
In this study, air quality measurements were obtained from Istanbul Metropolitan Municipality
Environment Protection and Control Office, a municipal agency with ten automatic air quality gauge
stations that measure air pollution in Istanbul's atmosphere. An 1-hour interval has been used to observe
these measurements. The variables utilized in the study are Ozone (μg/m3), Sulfur Dioxide (μg/m3), Nitric
Oxide(μg/m3), Nitrogen Dioxide (μg/m3), Dust (μg/m3), Total Hydrocarbon (μg/m3), Outdoor
Temperature (°𝐶), Wind Speed (m/s), Solar Irradiance (Hour), Cloudiness(0–10), Pressure (mbar), Relative
Humidity (%), Rain (mm). In order to capture multivariate correlations between these features of linear
values, multivariate groups are formed by combining these features of linear values. This multivariate
analysis reveals the importance of variables to capture high-quality multi-features in the predictionn. The
analysis shows when 𝑂3, THC, NO features are in the multivariate groups, these features are the best to
identify air pollution quality with .7658 accuracy rate. OT, SI, 𝑆𝑂2 comes after with ,7442 accuracy rate.
This study shows that by selecting only the most important features, we can reduce the complexity of the
model and improve its accuracy and generalization ability. Feature ranking can help us understand the
relationships between variables in a dataset and identify patterns or trends that may be important for
understanding the data. Feature ranking can also improve the interpretability of machine learning models.
Unlike previous studies, multi-feature comparison as a group instead of individual feature comparison
contributed to the reduction in data size. There was also a 17% improvement in calculation times which
revealed the importance of the study.