Prediksi Delay Time Penerbangan Pendekatan Model Lstm Dan Bilstm Dengan Interpretasi Shap (Studi Kasus : Bandara Internasional Juanda Surabaya)

Hidayah, Ifayanti Rohmatul (2026) Prediksi Delay Time Penerbangan Pendekatan Model Lstm Dan Bilstm Dengan Interpretasi Shap (Studi Kasus : Bandara Internasional Juanda Surabaya). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003241022_Master_Thesis.pdf Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Keterlambatan penerbangan merupakan isu strategis dalam industri transportasi udara karena berdampak pada kepuasan penumpang, efisiensi operasional, dan kerugian finansial. Penelitian ini bertujuan membangun model Prediksi durasi keterlambatan penerbangan di Bandara Internasional Juanda Surabaya dengan menggunakan Long Short Term Memory (LSTM) dan Bidirectional LSTM (BiLSTM) serta mengintegrasikan metode interpretabilitas Shapley Additive Explanations (SHAP). Data yang digunakan berupa data sekunder yang mencakup variabel operasional penerbangan (jadwal, rute, maskapai, jenis penerbangan) dan variabel cuaca (suhu, curah hujan, kecepatan angin, tekanan udara) dari periode 1 Januari 2023 hingga 31 Oktober 2025. Tahapan penelitian meliputi preprocessing data, normalisasi Min - Max, pembagian data secara kronologis menjadi training, validasi, dan testing, serta perancangan model dengan variasi neuron, hidden layer, epoch, batch size, dan optimizer Adam. Kinerja model dievaluasi menggunakan MAE dan RMSE, dan model terbaik dianalisis menggunakan SHAP untuk mengidentifikasi variabel dominan penyebab keterlambatan, baik secara global maupun lokal. Hasil penelitian menunjukkan bahwa keterlambatan mendominasi operasi penerbangan, yaitu sekitar tiga perempat penerbangan tergolong delay, dengan distribusi yang bersifat zero-inflated dan menjulur ke kanan. Model LSTM terbaik (3 lapisan tersembunyi, 4 neuron) menghasilkan MAE sebesar 15,035 menit dan RMSE sebesar 17,889 menit pada data uji, sedikit lebih baik dibandingkan BiLSTM (MAE 15,328 menit; RMSE 17,990 menit), sehingga LSTM ditetapkan sebagai model terbaik. Karena setiap penerbangan diperlakukan sebagai satu observasi independen (regresi satu-input dengan panjang runtun T=1), kedua model cenderung menghasilkan prediksi yang mendekati nilai rata-rata, sehingga mampu menangkap tingkat keterlambatan umum tetapi belum menjangkau keterlambatan ekstrem. Interpretasi SHAP mengungkap bahwa faktor waktu (komponen bulan/musim dan jam jadwal) paling berpengaruh terhadap keterlambatan, disusul identitas maskapai (seperti Lion Air) serta keterkaitan dengan bandara dan rute, sedangkan faktor cuaca berpengaruh relatif kecil. Dengan demikian, meskipun daya prediksi model masih terbatas, model yang dihasilkan tetap memberikan interpretasi yang transparan sehingga dapat menjadi masukan strategis bagi pengelola bandara dan maskapai dalam meningkatkan ketepatan waktu penerbangan
====================================================================================================================================
Flight delays are a strategic issue in the air transportation industry as they affect passenger satisfaction, operational efficiency, and financial performance. This study aims to develop a predictive model for flight delay duration at Juanda International Airport, Surabaya, using Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM), integrated with the Shapley Additive Explanations (SHAP) interpretability method. The data consist of secondary operational and weather data comprising flight operational variables (schedule, route, airline, and flight type) and weather variables (temperature, rainfall, wind speed, and air pressure) covering the period from January 2023 to October 2025. The research stages include data preprocessing, Min-Max normalization, chronological splitting into training, validation, and testing datasets, and model design with variations in the number of neurons, hidden layers, epochs, batch size, and the Adam optimizer. Model performance is evaluated using MAE and RMSE, and the best-performing model is further analyzed with SHAP to identify the dominant variables contributing to delays, both globally and locally. The results show that delays dominate flight operations, with about three-quarters of all flights delayed and a delay distribution that is zero-inflated and right-skewed. The best LSTM model (three hidden layers and four neurons) achieves an MAE of 15,035 minutes and an RMSE of 17,889 minutes on the testing set, slightly outperforming BiLSTM (MAE of 15,328 minutes and RMSE of 17,990 minutes). LSTM is therefore selected as the best model. Because each flight is treated as an independent observation (single-input regression with a sequence length of T=1), both models tend to predict close to the mean, capturing the general delay level but not extreme delays. The SHAP analysis reveals that temporal factors (the month/seasonal and scheduled-time components) are the most influential in predicting delays, followed by airline identity (such as Lion Air) and the association with airports and routes, whereas weather factors have a relatively minor effect. Thus, although the model's predictive power remains limited, it still provides a transparent interpretation, offering strategic insights for airport management and airlines to improve flight punctuality.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	BiLSTM, LSTM Prediksi durasi keterlambatan, SHAP BiLSTM, Flight Delay Time, LSTM, SHAP.
Subjects:	Q Science Q Science > Q Science (General) Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User:	Ifayanti Rohmatul Hidayah
Date Deposited:	04 Aug 2026 07:33
Last Modified:	04 Aug 2026 07:33
URI:	http://repository.its.ac.id/id/eprint/143226

Actions (login required)

View Item