Penerapan Symmetric Mean Absolute Percentage Error Terhadap Hasil Prediksi Multiple Linear Regression Dan Extreme Gradient Boosting Regression

Abstract

Banyaknya metode Machine learning dalam prediksi menyebabkan pemilihan algoritma menjadi penting karena setiap metode memiliki kemampuan berbeda dalam mengolah karakteristik data. Perbedaan performa antar metode menunjukkan perlunya pengujian untuk menentukan algoritma yang sesuai dalam menghasilkan prediksi akurat. Penelitian ini menggunakan data berupa jumlah penumpang kereta priority periode 2021–2025 dengan variabel pendukung berupa load factor, high season, indeks Google Trends, dan tingkat inflasi. Pengujian dilakukan dengan menggunakan metode 5-Fold Cross Validation. Sedangkan untuk memprediksi jumlah penumpang kereta tipe priority menggunakan metode Multiple Linear Regression (MLR) dan Extreme Gradient Boosting Regression (XGBoost Regression) dan evaluasi hasil prediksi menggunakan Symmetric Mean Absolute Percentage Error (sMAPE). Hasil penelitian menunjukkan bahwa MLR memiliki performa prediksi lebih baik dibandingkan XGBoost Regression pada kasus memprediksi jumlah penumpang kereta tipe priority. MLR memiliki selisih terhadap nilai aktual sebesar 1,387033912, sedangkan XGBoost Regression memiliki selisih sebesar 6.829,9905. Nilai rata-rata sMAPE metode MLR sebesar 0,0226%, sedangkan metode XGBoost Regression sebesar 16,5854%. Hasil tersebut menunjukkan MLR memiliki tingkat kesalahan lebih rendah dan lebih sesuai digunakan pada kasus ini.

The large number of Machine learning methods available for prediction makes algorithm selection important because each method has different capabilities in processing data characteristics. Differences in performance among methods indicate the need for testing to determine the appropriate algorithm for producing accurate predictions. This study uses data on the number of priority train passengers during the 2021–2025 period, with supporting variables including load factor, high season, Google Trends index, and inflation rate. Model testing was conducted using the 5-Fold Cross Validation method. The prediction of the number of priority train passengers was performed using Multiple Linear Regression (MLR) and Extreme Gradient Boosting Regression (XGBoost Regression), while the prediction results were evaluated using Symmetric Mean Absolute Percentage Error (sMAPE). The results showed that MLR achieved better prediction performance than XGBoost Regression in predicting the number of priority train passengers. MLR had a difference from the actual value of 1.387033912, while XGBoost Regression had a difference of 6,829.9905. The average sMAPE value of the MLR method was 0.0226%, whereas the XGBoost Regression method was 16.5854%. These results indicate that MLR has a lower error rate and is more suitable for use in this case.