Prediksi Return Emas menggunakan Algoritma Gradient Boosting dengan Optimasi PSO dan Pendekatan Shapley Additive Explanations (SHAP)

Nirwana, Yansen Demos Putra (2026) Prediksi Return Emas menggunakan Algoritma Gradient Boosting dengan Optimasi PSO dan Pendekatan Shapley Additive Explanations (SHAP). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221151-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Inovasi digital dan kemudahan akses teknologi telah mendorong peningkatan antusiasme masyarakat Indonesia untuk berpartisipasi dalam berbagai instrumen investasi. Saat ini, emas digital menjadi salah satu instrumen populer karena kemudahan transaksinya serta perannya yang krusial sebagai pelindung nilai terhadap tekanan inflasi dan pelemahan mata uang. Meskipun akses transaksinya semakin mudah, harga emas bergerak sangat fluktuatif karena sensitivitasnya terhadap ketidakpastian geopolitik, kepanikan pasar, dan dinamika kebijakan moneter global. Kompleksitas volatilitas yang ekstrem ini menuntut adanya pendekatan machine learning yang adaptif dan mampu memetakan pola non-linear secara akurat. Untuk menjawab tantangan tersebut, penelitian ini membandingkan XGBoost dan CatBoost untuk memprediksi return harga emas (XAU/IDR) menggunakan 571 observasi mingguan periode 7 Juni 2015 – 10 Mei 2026. Fitur lag diseleksi melalui PACF dan CCF, kemudian divalidasi ulang berdasarkan korelasi. Perbandingan model dilakukan dengan data final, dan model terbaik dioptimasi menggunakan Particle Swarm Optimization (PSO) serta diinterpretasikan dengan Shapley Additive Explanations (SHAP). Hasil pemodelan awal menunjukkan overfitting pada kedua model, namun model CatBoost memiliki generalisasi yang lebih tangguh dari XGBoost dengan nilai RMSE uji sebesar 1,71340. Optimasi PSO dengan skenario penurunan bobot inersia secara linier berhasil mengurangi overfitting dan meningkatkan kinerja CatBoost secara signifikan. PSO berhasil menurunkan RMSE uji menjadi 1,54258 dan menaikkan koefisien determinasi dari 0,58711 menjadi 0,66533. Analisis SHAP mengungkap bahwa pergerakan perak global (XAG/USD), XAU/IDR MA 3, dan nilai tukar Rupiah (USD/IDR), memberikan kontribusi positif dominan. Peningkatan volatilitas pasar (VIX) juga berkorelasi positif. Sebaliknya, Indeks Dolar AS (DXY) dan imbal hasil obligasi AS (US10Y) menekan harga akibat tingginya biaya peluang. Di sisi lain, return indeks saham S&P 500 lag 1 memiliki pengaruh global yang minim, namun menjalankan fungsi spesifik mendeteksi lonjakan volatilitas ekstrem pada titik waktu tertentu.
=======================================================================================================================================
Digital innovation and easy access to technology have fueled growing enthusiasm among Indonesians to participate in various investments. Currently digital gold is a popular instrument due to easy transactions and its crucial role as a store of value against inflationary pressures and currency depreciation. Despite easier access gold prices remain highly volatile due to their sensitivity towards geopolitical uncertainty as well as market panic and global monetary policy dynamics. The complexity of this extreme volatility demands an adaptive machine learning approach capable of accurately mapping non-linear patterns. To address this challenge this study compares XGBoost and CatBoost to predict gold price returns (XAU/IDR) using weekly observations from June 2015 to March 2022. Lagged features were selected via PACF and CCF before being revalidated based on correlation. Model comparison was conducted using the final dataset where the best model was optimized using Particle Swarm Optimization (PSO) and interpreted with Shapley Additive Explanations (SHAP). Initial modeling indicated overfitting in both models although CatBoost demonstrated stronger generalization than XGBoost with a test RMSE of 1.71340. PSO optimization with linearly decreasing inertia weights successfully reduced overfitting and significantly improved CatBoost’s performance. The algorithm successfully lowered the test RMSE to 1.54258 and increased the coefficient of determination from 0.58711 to 0.66533. SHAP analysis revealed dominant positive contributions from global silver prices (XAG/USD) and the 3-week moving average of gold (XAU/IDR MA 3) alongside the Rupiah exchange rate (USD/IDR). Increased market volatility (VIX) also showed a positive correlation. Conversely the U.S. Dollar Index (DXY) and U.S. bond yields (US10Y) exerted downward pressure on prices due to high opportunity costs. Meanwhile the 1-period lagged S&P 500 stock index return had minimal overall influence but served a specific function in detecting extreme volatility spikes at specific times.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Emas, Gradient Boosting, Particle Swarm Optimization, Prediksi, SHAP, Gold, Gradient Boosting, Particle Swarm Optimization, Prediction, SHAP
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HB Economic Theory > Economic forecasting--Mathematical models. Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > Q Science (General) > Q337.3 Swarm intelligence Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Yansen Demos Putra Nirwana
Date Deposited:	30 Jul 2026 03:15
Last Modified:	30 Jul 2026 03:15
URI:	http://repository.its.ac.id/id/eprint/140183

Actions (login required)

View Item