Academic Journal
Prediksi Inflasi Menggunakan Deep Learning dengan Integrasi Variabel Makroekonomi dan Data Narasi Digital sebagai Variabel Eksogen.
| Τίτλος: | Prediksi Inflasi Menggunakan Deep Learning dengan Integrasi Variabel Makroekonomi dan Data Narasi Digital sebagai Variabel Eksogen. (Indonesian) |
|---|---|
| Συγγραφείς: | Damayanti, Riza, Irhamah, Widhianingsih, Tintrim Dwi Ary |
| Πηγή: | Jurnal Locus: Penelitian dan Pengabdian; jul2026, Vol. 5 Issue 7, p5115-5125, 11p |
| Θεματικοί όροι: | Inflation forecasting, Deep learning, Dummy variables, Recurrent neural networks, Macroeconomics, Digital storytelling, Time series analysis |
| Γεωγραφικοί όροι: | Indonesia |
| Abstract (English): | This research extends the authors' prior work a six-model LSTM ablation research for Indonesian inflation forecasting (manuscript submitted to AMLDS 2026, under review) — by evaluating whether structural break dummy variables (Lebaran, COVID-19, fuel price hike) enhance models integrating macroeconomic and digital narrative variables. Using the same dataset (January 2005- December 2024, 240 monthly observations), this paper introduces M3_Naratif_Dummy as the new model contribution, combining Google Trends Index and IEH Bank Indonesia with three structural break dummies. Seven models are compared through ablation study: ARIMA, VAR, M1 LSTM Baseline, M2 LSTM Numerical, M3 LSTM Narrative, M3 LSTM Narrative+Dummy, and M4 LSTM Full. Hyperparameters are selected via sequential grid search (N=10 replications), yielding T*=6. Evaluation on test set 2020-2024 (60 months) shows M3 LSTM Narrative remains the best model (RMSE=0.7823, sMAPE=17.72%, std=0.53%), while M3_Naratif_Dummy achieves sMAPE=22.98% with 8.4x higher variance (std=4.43%). At optimal T=6, both models converge to near-identical performance (sMAPE 17.37% vs 17.57%), and Diebold-Mariano test confirms no significant difference (p=0.845). All LSTM models significantly outperform VAR (p<0.05). Key finding: digital narrative variables implicitly capture structural break information, making explicit dummy variables redundant and destabilizing for LSTM inflation forecasting. [ABSTRACT FROM AUTHOR] |
| Abstract (Indonesian): | Penelitian ini memperluas studi pendahuluan penulis ablation study enam model LSTM untuk peramalan inflasi Indonesia (manuscript submitted ke AMLDS 2026, under review) dengan mengevaluasi apakah variabel dummy structural break (Lebaran, COVID-19, kenaikan BBM) meningkatkan performa model yang mengintegrasikan variabel makroekonomi dan naratif digital. Menggunakan dataset yang sama (Januari 2005-Desember 2024, 240 observasi bulanan), penelitian ini memperkenalkan M3_Naratif_Dummy sebagai model baru yang mengombinasikan Google Trends Index dan IEH Bank Indonesia dengan tiga dummy structural break. Tujuh model dibandingkan: ARIMA, VAR, M1 LSTM Baseline, M2 LSTM Numerik, M3 LSTM Naratif, M3 LSTM Naratif+Dummy, dan M4 LSTM Full (N=10 replikasi, T*=6). Hasil menunjukkan M3 LSTM Naratif tetap menjadi model terbaik (RMSE=0,7823, sMAPE=17,72%, std=0,53%), sementara M3_Naratif_Dummy mencapai sMAPE=22,98% dengan varian 8,4 kali lebih tinggi (std=4,43%). Pada lookback optimal T=6, kedua model konvergen ke performa hampir setara (sMAPE 17,37% vs 17,57%), dan uji Diebold-Mariano mengonfirmasi tidak ada perbedaan signifikan (p=0,845). Seluruh model LSTM secara signifikan mengungguli VAR (p<0,05). Temuan utama: variabel naratif digital secara implisit menangkap informasi structural break, sehingga penambahan dummy eksplisit bersifat redundan dan mendestabilisasi model LSTM peramalan inflasi Indonesia. [ABSTRACT FROM AUTHOR] |
| Copyright of Jurnal Locus: Penelitian dan Pengabdian is the property of Riviera Publishing and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Βάση Δεδομένων: | Complementary Index |
καταχωρήστε σχόλιο πρώτοι!