Interpretable Machine Learning Dengan Pendekatan Model Agnostik Pada Prediksi Fuel Consumption Rate Mining Haul Truck

Arbianto, Domy Guruh Dwi (2025) Interpretable Machine Learning Dengan Pendekatan Model Agnostik Pada Prediksi Fuel Consumption Rate Mining Haul Truck. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032232019-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Model machine learning (ML) sering disebut sebagai “Black-Box” karena kerumitannya sehingga sulit diinterpretasikan oleh manusia. Interpretabilitas model menjadi sangat penting untuk memahami penyebab sebuah prediksi tertentu dibuat. Salah satunya dalam memahami perilaku dan menjelaskan pola fuel consumption rate (FCR) dari Mining Haul Truck menggunakan model prediksi. Seorang engineer akan kesulitan untuk menentukan kontributor paling signifikan secara mudah dan cepat. Pada sebuah prediksi, sebuah model ML tidak akan memberi tahu bagaimana sampai pada sebuah keputusan. Hal ini akan menimbulkan kebingungan engineer pada saat akan menganalisa kondisi dan menentukan prioritas perbaikan yang diperlukan. Prioritisasi perbaikan perlu dilakukan karena pertimbangan biaya, tingkat kesulitan, dan downtime yang sangat mempengaruhi produktivitas. Oleh karena itu, pendekatan model agnostik perlu dilakukan. Penelitian ini bertujuan untuk menghasilkan model prediksi untuk memahami perilaku dan pola FCR kemudian menginterpretasikannya dengan model agnostik. Penelitian ini menggunakan data Vehicle Health Monitoring System (VHMS) dari Agustus 2024 hingga Februari 2025 dengan penanganan outlier dan multikolinieritas. Algoritma machine learning yang digunakan adalah Random Forest Regressor (RFR) dan XGBoost Regressor (XGB). Model agnostik yang digunakan untuk interpretasi global adalah Partial Dependence Plot, Feature Interaction, dan Permutation Feature Importance. Sedangkan interpretasi lokal menggunakan Local Interpretable Model-Agnostic Explanations dan Shapley Value. Hasil evaluasi performa model menunjukkan bahwa tidak ada perbedaan akurasi model yang signifikan antara RFR maupun XGB. Model prediksi terbaik adalah model RFR Tanpa Normalisasi – Tanpa Penanganan Outlier dengan nilai RMSE 3,7312, SMAPE 4,64%, dan R-Squared 0,7936. Hasil interpretasi global dan lokal menunjukkan bahwa top three faktor yang berkontribusi signifikan terhadap FCR adalah engine speed, road angle, dan boost pressure.
====================================================================================================================================
Machine learning (ML) models are often referred to as "black boxes" because their complexity makes them difficult for humans to interpret. Model interpretability is crucial for understanding the causes of certain predictions. One example is understanding the behavior and explaining the fuel consumption rate (FCR) patterns of Mining Haul Trucks using predictive models. An engineer would struggle to easily and quickly determine the most significant contributors. In a prediction, an ML model does not provide information on how to arrive at a decision. This creates confusion for engineers when analyzing conditions and prioritizing necessary repairs. Prioritizing repairs is necessary because cost, difficulty, and downtime significantly impact productivity. Therefore, a model-agnostic approach is necessary. This research aims to develop a predictive model to understand FCR behavior and patterns and then interpret them using a model-agnostic approach. This study used Vehicle Health Monitoring System (VHMS) data from August 2024 to February 2025, with outlier and multicollinearity management. The machine learning algorithms used were Random Forest Regressor (RFR) and XGBoost Regressor (XGB). The agnostic models used for global interpretation were Partial Dependence Plot, Feature Interaction, and Permutation Feature Importance. Meanwhile, local interpretation used Local Interpretable Model-Agnostic Explanations and Shapley Values. The model performance evaluation results showed no significant difference in model accuracy between RFR and XGB. The best prediction model was the RFR Without Normalization – Without Outlier Handling model with an RMSE of 3.7312, SMAPE of 4.64%, and R-Squared of 0.7936. The results of global and local interpretations indicated that the top three factors significantly contributing to FCR were engine speed, road angle, and boost pressure.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Fuel Consumption Rate, Interpretable Machine Learning, Model Agnostik, Agnostic Model, Random Forest, XGBoost
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.8 Productivity. Efficiency
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Domy Guruh Dwi Arbianto
Date Deposited:	02 Dec 2025 05:05
Last Modified:	02 Dec 2025 05:05
URI:	http://repository.its.ac.id/id/eprint/128840

Actions (login required)

View Item