Implementasi Algoritma Random Forest Terhadap Prediksi Harga Bitcoin

Nova Fitriawan Zuhri (Universitas Islam Lamongan)
Nur Nafi’iyah (Universitas Islam Lamongan)

Jurnal Nasional Komputasi dan Teknologi Informasi

Abstrak - Tingginya volatilitas Bitcoin mendorong kebutuhan model prediktif presisi untuk landasan keputusan investasi optimal. Studi ini mengimplementasikan algoritma Random Forest guna memprediksi pergerakan harga berdasarkan 1.766 data historis harian (Januari 2020-Oktober 2024). Pra-pemodelan diawali analisis korelasi Pearson dengan ambang batas 0.5, yang menyeleksi fitur High (0,999), Open (0,998), dan Low (0,997) sebagai prediktor akibat asosiasi kuat, sementara Volume dan Change% dieliminasi karena kontribusi minimal. Pengujian membandingkan dua strategi pembagian data: partisi acak dan tidak acak (rasio 80:20), menggunakan metrik Mean Absolute Percentage Error (MAPE) dan Akurasi. Hasil empiris menunjukkan partisi acak unggul (MAPE 1,34%; Akurasi 98,66%) dibanding partisi tidak acak (MAPE 1,7%; Akurasi 98,3%). Konklusi menegaskan efektivitas signifikan algoritma Random Forest, dengan keberhasilan bergantung pada ketepatan seleksi fitur dan adaptasi strategi pembagian data terhadap karakteristik dataset.Kata kunci: Bitcoin; Random Forest; Prediksi Harga; Korelasi Pearson; MAPE; Abstract - The high volatility of Bitcoin necessitates precise predictive models for optimal investment decision-making. This study implements the Random Forest algorithm to predict price movements based on 1,766 daily historical data points (January 2020 - October 2024). Pre-modeling began with Pearson correlation analysis with a threshold 0.5, which selected the High (0.999), Open (0.998), and Low (0.997) features as predictors due to their strong association, while Volume and Change% were eliminated due to their minimal contribution. The testing compared two data splitting strategies: random and non-random (80:20 ratio), using the Mean Absolute Percentage Error (MAPE) and Accuracy metrics. Empirical results showed that random partitioning outperformed non-random partitioning (MAPE 1.34%; Accuracy 98.66% compared to MAPE 1.7%; Accuracy 98.3%). The conclusion confirms the significant effectiveness of the Random Forest algorithm, with success depending on the accuracy of feature selection and adapting the data splitting strategy to the characteristics of the dataset. Keywords: Bitcoin; Random Forest; Price Prediction; Pearson Correlation; MAPE;