Analisis Anomali Biaya Operasional Coal Getting dan Overburden Removal pada Industri Pertambangan menggunakan Machine Learning

Pranata, Okta Robian (2026) Analisis Anomali Biaya Operasional Coal Getting dan Overburden Removal pada Industri Pertambangan menggunakan Machine Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032242068-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Biaya operasional kegiatan coal getting dan overburden removal merupakan komponen terbesar di industri pertambangan batubara, namun sering tidak berbanding lurus dengan produktivitas. Variasi biaya yang sulit dipetakan secara manual, tingginya losstime, rendahnya utilisasi alat (rata-rata 44,3%), serta deviasi anggaran yang mencapai -12,4% pada bulan tertentu menjadi permasalahan utama yang dihadapi perusahaan. Penelitian ini menganalisis anomali biaya operasional CG dan OBR menggunakan pendekatan machine learning berbasis integrasi lima sumber data: ERP Ellipse, FMS CISEA, Equipment Monitoring System, Logbook Supervisor, dan RKAP Bulanan. Dataset terdiri dari 1.508 observasi harian valid periode Juli-Desember 2025 dengan 71 variabel. Dua metode machine learning diterapkan secara terintegrasi: (1) deteksi anomali unsupervised menggunakan ensemble Isolation Forest (IF) dan Local Outlier Factor (LOF), serta (2) prediksi biaya supervised menggunakan Multiple Linear Regression (MLR) dilengkapi analisis interpretabilitas SHAP. Evaluasi model menggunakan weak ground truth hasil validasi tiga expert praktisi pertambangan di perusahaan. Hasil penelitian menunjukkan struktur biaya didominasi komponen angkut sebesar 65,3% dari total Rp 1.608,4 miliar dengan rata-rata HPP Rp 43.559/BCM. Model ensemble IF+LOF dengan Ensemble Score P95 mengidentifikasi 76 observasi anomali (5,0%) dalam 11 tipologi (coverage 96,1%); tipologi Deviasi RKAP Ekstrem - Budget Driven paling dominan (17 kasus, 22,4%), dengan 16 observasi Kritikal P99 (1,06%) memerlukan investigasi prioritas. ELEKTRIFIKASI PTBA mencatat tingkat anomali tertinggi sebesar 8,60% dibandingkan kontraktor SPPH 95-PPA (2,85%) dan SPPH 17443-PPA (1,64%). Model MLR menghasilkan R²=0,9380 (testing set), MAE=Rp 137,65 juta, K-Fold CV R²=0,9368±0,0250. Analisis SHAP mengkonfirmasi Z4_Indeks_Efisiensi sebagai prediktor dominan (Std Beta=+0,8089), sedangkan breakdown alat (7,57×), losstime (2,55×), dan HPP/BCM (2,33×) merupakan faktor penyebab anomali paling signifikan (p<0,001). Seluruh hasil diwujudkan dalam dashboard analitik Streamlit empat modul (Descriptive, Diagnostic, Predictive, Executive Summary) yang mempersingkat siklus analisis. Penelitian ini menyimpulkan bahwa pendekatan machine learning efektif mendeteksi dan mengklasifikasikan anomali biaya operasional pertambangan, segingga peneliti dapat menghasilkan rekomendasi manajerial berbasis bukti untuk peningkatan efisiensi operasional berkelanjutan.
=======================================================================================================================================
Operational costs in coal getting (CG) and overburden removal (OBR) represent the largest expenditure in coal mining, yet they are often disproportionate to productivity. Cost variations that are difficult to map manually, high losstime, low equipment utilization (averaging 44.3%), and budget deviations reaching -12.4% in certain months are key challenges faced by the company. This study analyzes operational cost anomalies in CG and OBR activities using a machine learning approach integrating five data sources: ERP Ellipse, FMS CISEA, Equipment Monitoring System, Supervisor Logbooks, and Monthly RKAP. The dataset comprises 1,508 valid daily observations over July-December 2025 with 71 variables. Two machine learning methods are applied in an integrated pipeline: (1) unsupervised anomaly detection using an ensemble of Isolation Forest (IF) and Local Outlier Factor (LOF), and (2) supervised cost prediction using Multiple Linear Regression (MLR) with SHAP interpretability analysis. Model evaluation employs a weak ground truth constructed through structured validation by three mining practitioners at company. Results show that cost structure is dominated by hauling at 65.3% of total Rp 1,608.4 billion, with an average HPP of Rp 43,559 per BCM. The IF+LOF ensemble with Ensemble Score P95 threshold identified 76 anomalous observations (5.0%) across 11 typologies (96.1% coverage); the Extreme RKAP Deviation – Budget Driven typology is the most dominant (17 cases, 22.4%), with 16 Critical P99 observations (1.06%) requiring priority investigation. ELEKTRIFIKASI PTBA recorded the highest anomaly rate at 8.60% than SPPH 95-PPA (2.85%) and SPPH 17443-PPA (1.64%). The MLR model achieved R²=0.9380 (testing set), MAE=Rp 137.65 million, and K-Fold CV R²=0.9368±0.025. SHAP analysis identifies Z4_Indeks_Efisiensi as the primary cost predictor (Standardized Beta=+0.8089), while equipment breakdown (7.57×), losstime (2.55×), and HPP/BCM (2.33×) are the most significant anomaly causal factors (p<0.001). Findings are operationalized through a four-module Streamlit analytical dashboard (Descriptive, Diagnostic, Predictive, Executive Summary) that reduces the cost analysis cycle from 3–5 days to under 30 minutes. This study concludes that an machine learning approach effectively detects and classifies operational cost anomalies in mining, so this research can generating evidence-based managerial recommendations for continuous operational efficiency improvement.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Anomali Biaya, Coal Getting, Overburden Removal, Isolation Forest, Local Outlier Factor, Multiple Linear Regression, SHAP : Cost Anomaly, Coal Getting, Overburden Removal, Isolation Forest, Local Outlier Factor, Multiple Linear Regression, SHAP
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Okta Robian Pranata
Date Deposited:	20 Jul 2026 04:03
Last Modified:	20 Jul 2026 04:03
URI:	http://repository.its.ac.id/id/eprint/135576

Actions (login required)

View Item