Kusherawati, Ermi (2025) Pengembangan Model Prediksi Kelelahan Pekerja Turn Around Dengan Pendekatan Machine Learning : Studi Kasus PT Petrokimia Gresik. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6010241096-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6010241096-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy |
Abstract
Kelelahan kerja merupakan tantangan utama dalam lingkungan industri berisiko tinggi seperti konstruksi, manufaktur, dan pertambangan, di mana penurunan kesiapan kerja (fit to work) dapat berdampak signifikan terhadap keselamatan dan produktivitas. Penelitian ini mengembangkan model prediksi kesiapan kerja menggunakan pendekatan machine learning berbasis data fisiologis, subjektif, dan kognitif. Data dikumpulkan secara sistematis melalui pengukuran Heart Rate (HR), kuesioner Multidimensional Fatigue Inventory (MFI-20), dan Psychomotor Vigilance Test (PVT-3) dari partisipan yang menjalani aktivitas intensif selama Turnaround (TA). Analisis statistik menunjukkan bahwa tingkat kelelahan mental partisipan meningkat secara signifikan setelah bekerja, dengan nilai p=0,000 dan R² sebesar 91,65%, mengindikasikan pengaruh waktu kerja terhadap fluktuasi kelelahan. Model machine learning dengan algoritma Random Forest menunjukkan performa prediksi paling tinggi terhadap skor MFI (R²=0,999; RMSE=0,065) dan cukup akurat dalam memprediksi HR (R²=0,800; RMSE=127,109), menjadikannya model unggulan dalam proses evaluasi. Pendekatan ini memungkinkan pemilihan instrumen yang lebih efisien tanpa harus mengandalkan seluruh kombinasi alat ukur. Hasil klasifikasi kesiapan kerja membagi partisipan ke dalam tiga kategori: Fit to Work, Fit with Note, dan Temporary Unfit, berdasarkan data HR, MFI, dan waktu reaksi. Sebagian besar partisipan berada pada kategori Temporary Unfit, menunjukkan akumulasi kelelahan yang menghambat kesiapan kerja optimal. Implementasi tiga algoritma utama Random Forest, Support Vector Machine, dan Linear Regression membuktikan bahwa integrasi metode machine learning dapat menghasilkan prediksi kelelahan yang akurat, sekaligus mendukung pengambilan keputusan strategis dalam manajemen risiko keselamatan kerja.
=========================================================================================================================================
Work related fatigue poses a significant risk in high-demand industrial sectors such as construction, manufacturing, and mining, where reduced fitness for duty can directly impact safety and productivity. This research introduces a predictive model for work readiness by integrating machine learning techniques with physiological, subjective, and cognitive data sources. Systematic measurements were conducted using Heart Rate (HR) monitors, standardized fatigue questionnaires (MFI-20), and the Psychomotor Vigilance Test (PVT-3), collected across intensive work conditions during a Turnaround (TA) process. Statistical analysis revealed a consistent increase in mental fatigue over the course of the workday, with significant variation across time points (p = 0.000, R² = 91.65%). The Random Forest algorithm emerged as the most robust predictive model, demonstrating excellent performance in estimating MFI scores (R² = 0.999; RMSE = 0.065) and reliable accuracy in HR prediction (R² = 0.800; RMSE = 127.109). The findings suggest that MFI scores can be accurately predicted without requiring full instrumentation, thereby increasing data collection efficiency. Work readiness was classified into three categories: Fit to Work, Fit with Note, and Temporary Unfit, based on combined HR, MFI, and reaction time data. Most participants fell under the Temporary Unfit classification, indicating fatigue accumulation that impacted optimal readiness. Machine learning implementation using Random Forest, Support Vector Machine, and Linear Regression confirmed the potential of data-driven models in identifying fatigue levels and supporting safety-related decision making in high-risk environments.
Actions (login required)
 |
View Item |