The increasing use of the internet demands the availability of stable and high-quality network services. Changes in network traffic over each time period cause network speed to be dynamic, requiring an approach capable of predicting network speed based on historical patterns. This study aims to analyze the characteristics of Traffic Total Speed and predict internet network speed using the Random Forest Regression algorithm based on historical network traffic data at Universitas Jenderal Achmad Yani (Unjani). The research methodology includes Exploratory Data Analysis (EDA), preprocessing, feature engineering, lag feature construction, Random Forest Regression modeling, and model evaluation using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), coefficient of determination (R²), and 5-Fold TimeSeriesSplit. The initial dataset consisted of 722 observations, and after preprocessing and feature engineering, 716 observations were obtained. The data were then divided chronologically using a time-based split with an 80:20 ratio, consisting of 572 training data and 144 testing data. The testing results show that the model achieved an MAE of 121.03 Mbit/s, an MSE of 43,546.19 (Mbit/s)², an RMSE of 208.68 Mbit/s, a MAPE of 38.31%, and an R² of 0.8892. Evaluation using 5-Fold TimeSeriesSplit resulted in an average R² of 0.8049 ± 0.1178, an MAE of 171.86 ± 64.70 Mbit/s, and an RMSE of 286.25 ± 87.26 Mbit/s. The R² values for each fold were 0.5823, 0.8329, 0.8080, 0.8825, and 0.9188, respectively. Based on the feature importance analysis, Volume_lag1 was the most important feature with an importance score of 0.3512, followed by Speed_lag1 with a score of 0.1767. The results indicate that Random Forest Regression is capable of predicting Traffic Total Speed with reasonably good performance, although prediction errors remain higher during periods of significant changes in network speed, particularly during peak hours.
Copyrights © 2026