Pengaruh Transformasi Yeo–Johnson dan Standardisasi Data terhadap Kinerja Deep Neural Network dalam Prediksi Porositas Zona Produktif Berdasarkan Data Sumur

Widangsa, Alringga Rizky (2026) Pengaruh Transformasi Yeo–Johnson dan Standardisasi Data terhadap Kinerja Deep Neural Network dalam Prediksi Porositas Zona Produktif Berdasarkan Data Sumur. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025232018-Master_Thesis.pdf Restricted to Repository staff only Download (12MB)

Abstract

Porositas zona produktif merupakan salah satu parameter penting dalam karakterisasi reservoir karena menggambarkan kapasitas ruang pori batuan yang dapat menyimpan fluida. Prediksi porositas berdasarkan data sumur menghadapi tantangan berupa perbedaan skala antarfitur, distribusi yang tidak simetris, nilai nol, dan nilai ekstrem yang dapat memengaruhi proses pembelajaran Deep Neural Network (DNN). Penelitian ini menganalisis pengaruh transformasi Yeo–Johnson dan standardisasi terhadap kinerja DNN dalam memprediksi PORPAYX berdasarkan data sumur. Dataset yang digunakan terdiri atas 276 observasi dengan 24 fitur numerik sebagai variabel masukan dan PORPAYX sebagai variabel target. Model dibangun menggunakan arsitektur feed-forward Multilayer Perceptron dengan fungsi aktivasi ReLU pada lapisan tersembunyi dan satu neuron keluaran linear. Pengujian dilakukan melalui dua skenario, yaitu DNN tanpa preprocessing dan DNN dengan transformasi Yeo–Johnson yang dilanjutkan dengan standardisasi. Parameter preprocessing hanya dihitung dari data pelatihan pada setiap fold untuk mencegah data leakage. Optimasi hyperparameter dilakukan menggunakan Tree-structured Parzen Estimator melalui Optuna sebanyak 100 trial, sedangkan evaluasi menggunakan 5-fold cross-validation, R², MSE, RMSE, dan analisis residual. DNN terbaik selanjutnya dibandingkan dengan Linear Regression, Random Forest Regression, dan XGBoost Regression. DNN tanpa preprocessing menghasilkan MSE sebesar 0,0091 dan RMSE sebesar 0,0951 serta belum mampu menjelaskan variasi target dengan baik. Setelah transformasi Yeo–Johnson dan standardisasi diterapkan, model menghasilkan rata-rata R² sebesar 0,9477, MSE sebesar 0,0007, dan RMSE sebesar 0,0208. Hasil tersebut menunjukkan bahwa preprocessing meningkatkan akurasi dan kestabilan DNN secara substansial. Namun, Linear Regression dan Random Forest menghasilkan kinerja yang lebih tinggi pada dataset yang digunakan. Dengan demikian, transformasi Yeo–Johnson dan standardisasi efektif dalam meningkatkan kinerja DNN, meskipun penggunaan model yang lebih kompleks tidak selalu menghasilkan akurasi prediksi tertinggi.
===========================================================================================================================
Productive-zone porosity is an important parameter in reservoir characterization because it represents the pore-space capacity of reservoir rocks to store fluids. Predicting porosity from well data is challenging because the input variables may have different scales, asymmetric distributions, zero values, and extreme observations, all of which can affect the learning process of a Deep Neural Network (DNN). This study analyzes the effect of the Yeo–Johnson transformation and data standardization on DNN performance in predicting PORPAYX from well data. The dataset consists of 276 observations, with 24 numerical features used as input variables and PORPAYX as the target variable. The model was developed using a feed-forward Multilayer Perceptron architecture with ReLU activation in the hidden layer and a single linear output neuron. Two experimental scenarios were evaluated: a DNN without preprocessing and a DNN using the Yeo–Johnson transformation followed by standardization. The preprocessing parameters were fitted only to the training data in each fold to prevent data leakage. Hyperparameter optimization was performed using the Tree-structured Parzen Estimator through Optuna for 100 trials. Model performance was evaluated using 5-fold cross-validation, the coefficient of determination (R²), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and residual analysis. The best DNN scenario was also compared with Linear Regression, Random Forest Regression, and XGBoost Regression. The DNN without preprocessing produced an MSE of 0.0091 and an RMSE of 0.0951 and was unable to adequately explain the variation in the target variable. After applying the Yeo–Johnson transformation and standardization, the model achieved an average R² of 0.9477, an MSE of 0.0007, and an RMSE of 0.0208. These results demonstrate that preprocessing substantially improved the accuracy and stability of the DNN. However, Linear Regression and Random Forest Regression achieved higher overall performance on the dataset. Therefore, the Yeo–Johnson transformation and standardization were effective in improving DNN performance, although greater model complexity did not necessarily produce the highest predictive accuracy.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deep Neural Network, Multilayer Perceptron, Optuna, PORPAYX, porositas zona produktif, standardisasi, transformasi Yeo–Johnson, Deep Neural Networks, Yeo–Johnson, preprocessing, Tree-structured Parzen Estimator, well production prediction
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering
Depositing User:	Alringga Rizky Widangsa
Date Deposited:	03 Aug 2026 05:18
Last Modified:	03 Aug 2026 05:18
URI:	http://repository.its.ac.id/id/eprint/142011

Actions (login required)

View Item