Pengembangan Model Hybrid Arima-Machine Learning untuk Prediksi Harga Saham BCA

Abstrak

Penelitian ini bertujuan untuk menganalisis kinerja metode hybrid antara algoritma machine learning dan model ARIMA dalam memprediksi harga saham Bank BCA selama lima tahun terakhir. Data yang digunakan berasal dari saham Bank BCA periode 13 November 2019 hingga 12 November 2024, diperoleh melalui Yahoo Finance. Dataset ini terdiri dari 1210 record dengan tujuh variabel: Date, Open, Close, High, Low, Volume, dan Adj Close. Pengujian dilakukan memodelkan data linier menggunakan ARIMA, kemudian memprediksi residual menggunakan algoritma machine learning yaitu KNN, Naïve Bayes, Logistic Regression, SVM, Random Forest, dan Gradient Boost. Selanjutnya Prediksi Akhir didapatkan dari penjumlahan Prediksi ARIMA dengan Prediksi Residual oleh Machine Learning. Hasil evaluasi menunjukkan bahwa model hybrid ARIMA–SVM memberikan performa terbaik dengan nilai MSE sebesar 13.341,72, MAE sebesar 89,69, dan MAPE sebesar 0,9078%. Model ini juga memiliki nilai korelasi (R) tertinggi sebesar 0,9785. Sementara itu, model ARIMA–Gradient Boosting juga menunjukkan performa yang kompetitif dengan MSE sebesar 14.126,60 dan MAPE sebesar 0,9434%. Temuan ini menunjukkan bahwa pendekatan hybrid efektif dalam meningkatkan akurasi dan kestabilan prediksi saham, serta dapat dijadikan alternatif yang unggul dalam analisis pasar keuangan berbasis data historis.

Abstract

This study aims to analyze the performance of a hybrid method combining machine learning algorithms and the ARIMA model in predicting the stock prices of Bank BCA over the past five years. The data used were obtained from Yahoo Finance, covering the period from November 13, 2019, to November 12, 2024. The dataset consists of 1,210 records and includes seven variables: Date, Open, Close, High, Low, Volume, and Adjusted Close. The testing procedure involved modeling the linear component of the data using ARIMA, followed by predicting the residuals with machine learning algorithms, namely K-Nearest Neighbors (KNN), Naïve Bayes, Logistic Regression, Support Vector Machine (SVM), Random Forest, and Gradient Boosting. The final prediction was obtained by summing the ARIMA forecast with the residual predictions from the machine learning models. Evaluation results show that the hybrid ARIMA–SVM model delivered the best performance with an MSE of 13,341.72, MAE of 89.69, and MAPE of 0.9078%, along with the highest correlation (R) value of 0.9785. The ARIMA–Gradient Boosting model also demonstrated competitive performance with an MSE of 14,126.60 and a MAPE of 0.9434%. These findings indicate that the hybrid approach is effective in enhancing the accuracy and stability of stock price predictions and can serve as a promising alternative in historical data-based financial market analysis.