Skripsi

ANALISIS KOMPARATIF ROBUSTNESS ALGORITMA ENSEMBLE LEARNING TERHADAP PERGESERAN DISTRIBUSI DATA MUSIMAN PADA KASUS TRANSPORTASI CERDAS

---

Data distribution changes over time have become one of the main challenges in the application of machine learning to intelligent transportation systems. This phenomenon, known as concept drift, can lead to performance degradation when the testing data exhibit characteristics that differ from those of the training data. This study aims to analyze the robustness of machine learning algorithms against seasonal data distribution shifts (seasonal concept drift) in intelligent transportation applications. The dataset used consists of Chicago taxi trip records from 2024–2025, containing more than 13 million observations before preprocessing. Three machine learning algorithms were evaluated: Random Forest, XGBoost, and LightGBM. The evaluation was conducted using a cross-season approach by comparing model performance under baseline and cross-season scenarios. Model performance was measured using Mean Absolute Error (MAE), while distributional changes were analyzed using the Kolmogorov–Smirnov test and Spearman correlation analysis. The results show that seasonal concept drift affects the performance of all evaluated models. Under baseline testing, Random Forest achieved the best performance with MAE values ranging from 1.78 to 2.28, while XGBoost and LightGBM achieved MAE values ranging from 3.01 to 3.72 and 3.27 to 3.97, respectively. However, in cross-season testing, Random Forest experienced the highest average performance degradation of 71.37%, whereas XGBoost and LightGBM showed average degradations of 9.44% and 3.69%, x respectively. Correlation analysis revealed a positive relationship between data distribution shifts and model performance degradation. LightGBM demonstrated the highest level of robustness, exhibiting the lowest performance degradation and the weakest correlation with distributional changes. Furthermore, the supplementary comparative evaluation indicated that boosting-based models, particularly XGBoost and LightGBM, delivered more stable and consistent performance than Random Forest when facing seasonal distribution shifts. Keywords: Machine Learning, Robustness, Seasonal Concept Drift, Intelligent Transportation Systems, Taxi Trips.

---

Availability

Detail Information

Series Title

-

Call Number

T2013082026

Publisher

Indralaya : Prodi Sistem Komputer, Fakultas Ilmu Komputer Universitas Sriwijaya., 2026

Collation

xvii, 110 hlm.; ilus.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

006.310 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Sistem Komputer
Algoritma Ensemble Learning

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment