Publication: Increasing firm's efficiency with machine learning algorithms : an application in the logistics industry
Abstract
Hava kargo taşımacılığının hızla yaygınlaşması ile birlikte sektördeki problemlerin algoritmalar ve yazılımlar aracılığıyla çözülmesi kaçınılmaz hale gelmiştir. Bu çalışmada, hava kargo yük planlama problemi olarak adlandırılan hava kargo taşımacılığında karşılaşılan temel problemler ele alınmış ve literatürdeki çözüm yaklaşımları incelenmiştir. Kargo kapasite tahminleme problemine çözüm bulmak için dört farklı makine öğrenmesi modeli geliştirilmiştir. Temel amaç, yolcu uçuşlarında bagajlar için kullanılacak konteyner sayısını tahmin ederek geriye kalan kargo kapasitesini hesaplayabilmektir. Bu amaçla Lineer Regresyon, Karar Ağacı Regresyonu, Rassal Orman Regresyonu ve Ekstrem Gradyan Artırma Regresyonu modelleri kullanılarak geçmiş veriler üzerinden tahminler yapılmıştır. Veri seti, son 3 yılda gerçekleştirilen yaklaşık 130.000 yolcu uçuşu verisine sahip bir havayolu şirketinden elde edilmiştir. Tüm makine öğrenmesi modellerinin hiper parametreleri 5 kat çapraz doğrulama ile Rastgele Arama algoritması kullanılarak belirlenmiştir. Modeller, ortalama mutlak hata, ortalama hatanın karesi, ortalama hatanın karekökü ve R² puanı gibi istatistiksel metrikler kullanılarak birbirleriyle karşılaştırılmıştır. Dört farklı model arasından Karar Ağacı Regresyon modeli en iyi performansı göstermiştir. Seçilen model test senaryolarında oldukça başarılı sonuçlar vermiştir.
With the rapid spread of air cargo transportation, it has become inevitable to solve business problems in the sector by means of algorithms and software. In this study, the main problems encountered in air cargo transportation called as air cargo load planning problem are discussed and solution approaches in the literature are examined. In order to find a solution to the cargo capacity estimation problem, four different machine learning models were developed. The main objective is to predict number of containers to be used by baggages on passenger flights, so then the remaining cargo capacity can be calculated. In this purpose, Linear Regression, Decision Tree Regression, Random Forest Regression, and Extreme Gradient Boosting models were employed to make predictions from historical data. The dataset was gathered from an airline company having nearly 130.000 passenger flights data operated in the last 3 years. Hyperparameters of all machine learning models were determined using Random Search algorithm with 5-fold cross-validation. The models were compared with each other using statistical metrics such as mean absolute error, mean squared error, root mean squared error, and R² score. Among four different models, Decision Tree Regression model performed the best. In the test scenarios, the selected model produced successful results.
With the rapid spread of air cargo transportation, it has become inevitable to solve business problems in the sector by means of algorithms and software. In this study, the main problems encountered in air cargo transportation called as air cargo load planning problem are discussed and solution approaches in the literature are examined. In order to find a solution to the cargo capacity estimation problem, four different machine learning models were developed. The main objective is to predict number of containers to be used by baggages on passenger flights, so then the remaining cargo capacity can be calculated. In this purpose, Linear Regression, Decision Tree Regression, Random Forest Regression, and Extreme Gradient Boosting models were employed to make predictions from historical data. The dataset was gathered from an airline company having nearly 130.000 passenger flights data operated in the last 3 years. Hyperparameters of all machine learning models were determined using Random Search algorithm with 5-fold cross-validation. The models were compared with each other using statistical metrics such as mean absolute error, mean squared error, root mean squared error, and R² score. Among four different models, Decision Tree Regression model performed the best. In the test scenarios, the selected model produced successful results.