Atıf Sayısı
56
Görüntüleme
45
İndirme
5
Hava kirliliği, günümüzün en büyük sorunlarından birini teşkil etmektedir. Hava kirliliği, nüfusun artması, kentsel gelişme ve büyüme, endüstrinin gelişmesiyle giderek artan bir önem arz etmektedir. Genellikle hava kirleticilerinin insanlara, canlılara ve çevreye zararlı etkileri zaman, mekan, etki süresi, konsantrasyon ve diğer karakteristiklerine bağlı olarak karmaşık dağılım şekilleri göstermektedir. Bu karmaşıklık, kirletici örnekleri ve eğilimleri modelleme veya ölçmede, ayrıca insanların maruz kaldığı seviyeleri tahmin etmenin zor olduğu anlamına gelmektedir. Hava kirliliğinin önlenmesi konusunda yapılacak çalışmalar içerisinde en önemli adımlardan biri kirlenme olayının bir model içerisinde değerlendirilmesidir. Bu çalışmada Kastamonu ili ele alınarak, meteoroloji ve çevre uygulamalarında oldukça yeni ve başarılı sonuçlar veren çeşitli makine öğrenmesi algoritmaları ile hava kirliliğinin tahmininde, bazı meteorolojik parametreler kullanılarak hava kirliliği modellenmesi amaçlanmıştır. Normalizasyon tekniklerinin sistem performansına etkisini gözlemleyebilmek için veri setindeki değerler Minimum-Maksimum (Min-Max) normalizasyon teknikleri ile normalize edilmiştir. Çalışmada elde edilen performans değerleri, literatürdeki benzer çalışmalarla kıyaslanarak problemin çözümüne ilişkin en uygun sınıflandırma algoritması tespit edilmiştir. Yapay Sinir Ağları (YSA), Rastgele Orman (Random Forest), K-En Yakın Komşu (K-Nearest Neighborhood), Lojistik Regresyon (Logistic Regression), Karar Ağacı (Decision Tree), Lineer Regresyon (Linear Regression) ve Basit Bayes (Naive Bayes) gibi yöntemler kullanılarak ayrı ayrı modeller tasarlanmış ve sonuçlar analiz edilmiştir. Veri setinin %70’i eğitim ve %30’si test verisi olarak ayrılmıştır. Çalışma sonucunda, yapay sinir ağları beklendiği gibi %91 oranında başarı göstererek en doğru tahmini yapan yöntem olarak belirlenirken, sınıflandırmalardan en başarısız sonuç Lineer Regresyon ile %30 elde edilirken, en başarılı sonuçlar Rastgele Orman ve Karar Ağacı ile %99 elde edilmiştir. KastamonuDataSet üzerinde kullanılan yöntemlerin performans değerlendirmelerinde kullanılan temel performans göstergeleri olarak Açıklayıcılık Katsayısı (R2), Ortalama Karesel Hata (Mean Squared Error-MSE), Ortalama Hata Kare Kökü (Root Mean Square Error-RMSE) ve Ortalama Mutlak Hata (Mean Absolute Error-MAE) metrikleri bakımından istatistiksel önemli farklılıkların bulunduğu tespit edilmiştir.
Air pollution is one of the biggest problems today. Air pollution, the increase in population, urban development and growth have an increasing importance with the development of the industry. Usually, the harmful effects of air pollutants on humans, animals and the environment show complex forms of distribution depending on time, space, time, concentration and other characteristics. This complexity means that it is difficult to predict the levels that people are exposed to in modeling or measuring polluting samples and trends. One of the most important steps in the work on air pollution prevention is the assessment of the pollution incident within a model. In this study, Kastamonu has been dealt with, with various machine learning algorithms that give quite new and successful results in meteorology and environmental applications, with the air pollution forecast, with some meteorological parameters aimed at modeling air pollution. To monitor the impact of normalization techniques on the system performance, the values in the data set are normalized with the Minimum-Maximum (Min-Max) normalization techniques. The performance values obtained in the study, compared to similar studies in literature, have been identified as the most appropriate classification algorithm for the solution of the problem. Individual models were designed using methods such as artificial nerve networks (YSA), random forest (Random Forest), K-Nearest Neighborhood (K-Nearest Neighborhood), logistical regression (Logistic Regression), Decision Tree (Decision Tree), Linear Regression (Linear Regression) and Simple Bayes (Naive Bayes) and the results were analyzed. 70% of the data set is assigned to training and 30% to test data. The study found that artificial nerve networks were identified as the most accurate method of prediction with 91% success as expected, while the most unsuccessful outcome of classifications was obtained with 30% Linear Regression, the most successful outcome was obtained with 99% random Forest and Decision Tree. Basic performance indicators used in the performance assessment of the methods used on DataSet have been identified with statistical significant differences in terms of the Explanation Scorecard (R2), Mean Squared Error (MSE), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE).
Air pollution is one of the biggest problems of today. Air pollution, population growth, urban development and growth are increasingly important with the development of industry. Generally, the harmful effects of air pollutants on humans, animals and the environment show complex distribution patterns depending on time, space, duration of action, concentration and other characteristics. This complexity means that modeling and measurement of pollutant samples and trends is also difficult to predict the levels of pollution to which people are exposed. One of the most important steps in prevention of air pollution is the evaluation of contamination in a model. In this study, it is aimed to model air pollution by using some meteorological parameters in the estimation of air pollution by various machine learning algorithms which give new and successful results in meteorology and environment applications. Minimum-Max (Min-Max) normalization technique was used with learning methods. The performance values obtained in the study are compared with the similar studies in the literature and the most appropriate classification algorithm for the solution of the problem has been determined. Separate models were designed and analyzed by using methods such as Artificial Neural Networks (ANN), Random Forest, K-Nearest Neighborhood (K-NN), Logistic Regression, Decision Tree, Linear Regression and Naive Bayes. The performance values obtained in the study were compared with similar studies in the literature and the most appropriate estimation algorithm for the solution of the problem was determined. In this case, 70% of the data set is used for training and 30% for testing. As a result of the study, it was seen that the correct estimation rate for the ANN model was 87% and the other machine learning models gave the best results in the estimation with 99% of the Random Forest accuracy rate and 99% of the Decision Tree accuracy rate. The Linear Regression method performs poorly with a 30% accuracy rate. Performance evaluation of methods used on KastamonuDataSet in terms of the Explanatory Coefficient (R2), Mean Squared Error (MSE), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics.
Alan : Fen Bilimleri ve Matematik; Mühendislik
Dergi Türü : Uluslararası
Makale Detay 
Atıf Yapanlar
Benzer Makaleler
PDF Görüntüle
Dergi Bilgisi
Eseri Dinleyin
Bu Sayfayı Yazdırın
Paylaş
Mendeley
