Pemodelan Faktor-Faktor Yang Berpengaruh Pada Kualitas Semen Curah SprintPro Di PT Semen Indonesia (Persero) Tbk Menggunakan Support Vector Regression (SVR)

Rachman, Devi Sagita (2026) Pemodelan Faktor-Faktor Yang Berpengaruh Pada Kualitas Semen Curah SprintPro Di PT Semen Indonesia (Persero) Tbk Menggunakan Support Vector Regression (SVR). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221172-Undergaduate_thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Kualitas semen merupakan salah satu faktor penting yang menentukan keberhasilan konstruksi sehingga pengendalian mutu produk perlu dilakukan secara konsisten. Penelitian ini bertujuan untuk membangun model Support Vector Regression (SVR) dalam memodelkan hubungan antara karakteristik kimia dan fisika semen dengan kualitas semen curah SprintPro, mengevaluasi performa model berdasarkan nilai Mean Absolute Error (MAE), Root Mean Square Error (RMSE), dan koefisien determinasi (R²), menginterpretasikan model menggunakan SHAP, PDP dan ICE, serta menentukan kombinasi karakteristik kimia dan fisika semen yang optimal menggunakan pendekatan optimasi multiresponse berbasis desirability function. Data yang digunakan merupakan data sekunder hasil pengujian Laboratorium Product Quality Assurance PT Semen Indonesia (Persero) Tbk periode Januari 2023 hingga Desember 2024 dengan variabel prediktor berupa karakteristik kimia dan fisika semen serta variabel respon berupa kuat tekan dan setting time. Tahapan analisis meliputi pre-processing data, eksplorasi data, pembentukan model regresi linear sebagai baseline, pemodelan SVR, optimasi hyperparameter menggunakan Grid Search, seleksi fitur menggunakan Recursive Feature Elimination (RFE), interpretasi model, serta optimasi multiresponse menggunakan desirability function. Hasil penelitian menunjukkan bahwa model terbaik pada kelompok kuat tekan adalah SVR hasil RFE dengan hyperparameter C = 100, γ = 0,01, dan ε = 0,5, sedangkan model terbaik pada kelompok setting time adalah SVR hasil Grid Search dengan hyperparameter C = 50, γ = 0,01, dan ε = 0,5. Kedua model tersebut memberikan performa prediksi yang lebih baik dibandingkan model regresi linear maupun SVR dengan hyperparameter bawaan. Hasil optimasi multiresponse menghasilkan kombinasi karakteristik kimia dan fisika semen optimum dengan nilai overall desirability sebesar 0,8628, lebih tinggi dibandingkan kondisi best existing sebesar 0,5218. Dengan demikian, pendekatan SVR yang dipadukan dengan optimasi hyperparameter, seleksi fitur, interpretasi model, dan optimasi multiresponse berbasis desirability function mampu memodelkan kualitas semen curah SprintPro dengan baik serta menghasilkan rekomendasi kombinasi karakteristik kimia dan fisika semen yang berpotensi meningkatkan kualitas produk.
=========================================================================================================================================
Cement quality is one of the key factors determining the success of a construction project; therefore, consistent quality control of the product is essential. This study aims to develop a Support Vector Regression (SVR) model to model the relationship between the chemical and physical characteristics of cement and the quality of SprintPro bulk cement, evaluate the model’s performance based on the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and coefficient of determination (R²), interpret the model using SHAP, PDP, and ICE, and to determine the optimal combination of cement’s chemical and physical characteristics using a multiresponse optimization approach based on a desirability function. The data used consists of secondary data from tests conducted by the Product Quality Assurance Laboratory of PT Semen Indonesia (Persero) Tbk from January 2023 to December 2024, with predictor variables comprising the chemical and physical characteristics of cement and response variables comprising compressive strength and setting time. The analysis stages included data preprocessing, data exploration, building a linear regression model as a baseline, SVR modeling, hyperparameter optimization using Grid Search, feature selection using Recursive Feature Elimination (RFE), model interpretation, and multiresponse optimization using a desirability function. The results of the study show that the best model for the compressive strength group is the SVR model obtained via RFE with hyperparameters C = 100, γ = 0.01, and ε = 0.5, while the best model for the setting time group is the SVR model obtained via Grid Search with hyperparameters C = 50, γ = 0.01, and ε = 0.5. Both models provide better predictive performance compared to linear regression models or SVR models with default hyperparameters. The multiresponse optimization results yielded an optimal combination of cement’s chemical and physical characteristics with an overall desirability value of 0.8628, which is higher than the best existing condition of 0.5218. Thus, the SVR approach—combined with hyperparameter optimization, feature selection, model interpretation, and multiresponse optimization based on a desirability function—is capable of effectively modeling the quality of SprintPro bulk cement and generating recommendations for combinations of cement’s chemical and physical characteristics that have the potential to improve product quality.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Desirability Function, Kualitas Semen, Optimasi Multiresponse, Setting Time, Support Vector Regression, Kuat Tekan, Desirability Function, Cement Quality, Multiresponse Optimization, Compressive Strength
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA353.K47 Kernel functions (analysis) Q Science > QC Physics > QC173.4.C63 Composite materials
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Devi Sagita Rachman
Date Deposited:	03 Aug 2026 10:25
Last Modified:	03 Aug 2026 10:25
URI:	http://repository.its.ac.id/id/eprint/142383

Actions (login required)

View Item