Prediksi Pergerakan Harga Saham Bank Mandiri Menggunakan Metode Support Vector Regression dan Algoritma Grid Search

Samuelly, Francesco Totti (Universitas Kristen Satya Wacana)
Nataliani, Yessica (Universitas Kristen Satya Wacana)

Jutisi: Jurnal Ilmiah Teknik Informatika dan Sistem Informasi
Vol 14, No 3: Desember 2025 (Naskah Accepted)

The volatile nature of stock price movements poses a major challenge for investors in making accurate investment decisions. This study aims to predict the stock price movement of PT. Bank Mandiri (Persero) Tbk [BMRI] using Support Vector Regression (SVR) optimized through the Grid Search algorithm. The dataset consists of daily stock prices from August 2020 to August 2025, including open, high, low, close, adjusted close, and trading volume. The research process involves data collection, preprocessing (cleaning, feature selection, normalization), splitting into training and testing sets, parameter optimization using Grid Search with Leave-One-Out Cross Validation (LOOCV), model training, and evaluation with R², MSE, and RMSE. The results show that the SVR model with a linear kernel, C = 1 and epsilon = 0.01, achieved the best performance, with high accuracy (R² = 0.9991 on training data and R² = 0.9976 on testing data). These findings confirm the effectiveness of Grid Search–based SVR in predicting stock prices and supporting investment decision-making.Keywords: Stock Price Prediction; Support Vector Regression; Grid Search; Bank Mandiri AbstrakPergerakan harga saham yang fluktuatif menjadi tantangan utama bagi investor dalam menentukan strategi investasi yang tepat. Penelitian ini bertujuan memprediksi pergerakan harga saham PT. Bank Mandiri (Persero) Tbk [BMRI] dengan metode Support Vector Regression (SVR) yang dioptimalkan menggunakan algoritma Grid Search. Data yang digunakan berupa harga saham harian periode Agustus 2020–Agustus 2025, mencakup variabel open, high, low, close, adjusted close, dan volume. Tahapan penelitian meliputi pengumpulan data, pra-pemrosesan (pembersihan, seleksi fitur, normalisasi), pembagian data latih dan uji, optimasi parameter dengan Grid Search berbasis Leave-One-Out Cross Validation (LOOCV), pelatihan model, serta evaluasi dengan R², MSE, dan RMSE. Hasil penelitian menunjukkan SVR dengan kernel linear, parameter C = 1 dan epsilon = 0,01 memberikan performa terbaik dengan akurasi tinggi (R² = 0,9991 pada data latih dan R² = 0,9976 pada data uji). Temuan ini menegaskan efektivitas SVR berbasis Grid Search dalam memprediksi harga saham dan mendukung pengambilan keputusan investasi.