Headline Performance (Task 15 Final)
Performa Utama (Task 15 Final)
| Metric | Metrik | Value | Nilai | Relevance | Relevansi |
|---|---|---|---|---|---|
| Composite R² (103 periods, marginalised) | Composite R² (103 periode, marginalised) | 0.7091 | 0,7091 | Headline operational metric | Metrik operasional utama |
| Oracle upper bound (Stage-1 perfect) | Oracle upper bound (Stage-1 sempurna) | 0.7779 | 0,7779 | Stage-2 capacity ceiling | Plafon kapasitas Stage-2 |
| Stage-1 OOF balanced accuracy | Balanced accuracy OOF Stage-1 | 81.68% | 81,68% | Safety-relevant (minority Damaging class) | Relevan keselamatan (kelas Damaging minoritas) |
| Stage-0 URPD AUC (0.5 s window) | AUC URPD Stage-0 (window 0,5 s) | 0.9136 | 0,9136 | Sub-second near-field alert | Alert near-field sub-detik |
| Blind zone reduction | Reduksi blind zone | 38 → 11 km | 38 → 11 km | −71% with 100% Damaging Recall | −71% dengan 100% Damaging Recall |
| Golden Time Compliance | Golden Time Compliance | 99.44% | 99,44% | Traces with latency budget ≤ P-S | Trace dengan budget latensi ≤ P-S |
Pipeline Optimisation Journey
Perjalanan Optimisasi Pipeline
| Stage | Tahap | Method Added | Metode Ditambah | R² | Δ |
|---|---|---|---|---|---|
| task 11 | task 11 | Class-conditional marginalisation baseline | Baseline class-conditional marginalisation | 0.6074 | — |
| task 12 | task 12 | + Site features (VS30, log-distance) | + Fitur site (VS30, log-distance) | 0.6641 | +0.057 |
| task 14 | task 14 | + Dai multi-window (2, 5, 10 s post-P) | + Dai multi-window (2, 5, 10 s pasca-P) | 0.6966 | +0.033 |
| task 15 | task 15 | + XGB+LGB ensemble, SMOTE, Optuna, freq features | + ensemble XGB+LGB, SMOTE, Optuna, fitur freq | 0.7091 | +0.013 |
Total improvement: +0.102 across four feature-engineering generations under identical GroupKFold 5-fold CV protocol (seed 42).
Total peningkatan: +0,102 dalam empat generasi feature-engineering dengan protokol CV GroupKFold 5-fold identik (seed 42).
Site Sections
Seksi Situs
Final Operational Results
Hasil Operasional Final
Full Task 15 breakdown: composite R² = 0.7091 headline, 90-feature architecture, 4-stage pipeline summary, and retrospective event validation. The authoritative result page.
Rincian lengkap Task 15: composite R² = 0,7091, arsitektur 90-fitur, ringkasan pipeline 4-stage, dan validasi retrospektif event. Halaman hasil otoritatif.
CSV Cross-check (historical)
Cross-check CSV (historis)
Original audit identifying three mixed model runs in legacy reports/. Superseded by task 15's authoritative CSVs in reports/performance/ida_ptw_task15.*.
Audit awal mengidentifikasi tiga run model tercampur di reports/ legacy. Digantikan oleh CSV otoritatif task 15 di reports/performance/ida_ptw_task15.*.
Manuscript Audit (historical)
Audit Manuskrip (historis)
Original audit of manuscript_draft_IEEE.md. The current submission-ready version is manuscript_draft_IEEE_v2.md with integrated task-15 numbers.
Audit awal manuscript_draft_IEEE.md. Versi siap-submit saat ini adalah manuscript_draft_IEEE_v2.md dengan angka task-15 terintegrasi.
Evidence A–D Appendix
Lampiran Evidence A–D
One-page English appendix: Fixed-vs-IDA benchmark, information ceiling, saturation test, P-arrival sensitivity.
Lampiran satu halaman berbahasa Inggris: benchmark Fixed-vs-IDA, information ceiling, saturation test, sensitivitas P-arrival.
Multi-panel Figure
Figur Multi-panel
4-panel publication-ready: R² vs Period, anchor bar chart, residual distribution, gap-to-ceiling decomposition.
4-panel siap publikasi: R² vs Periode, bar chart anchor, distribusi residual, dekomposisi gap-to-ceiling.
Publication Readiness (historical)
Kesiapan Publikasi (historis)
Original Go/No-Go scorecard (April 2026). Status has since progressed through task 12 → 14 → 15; current state: ready for IEEE Access submission.
Scorecard Go/No-Go awal (April 2026). Status sejak itu berkembang melalui task 12 → 14 → 15; status saat ini: siap submission IEEE Access.
SIG-BMKG Analysis (ref)
Analisis SIG-BMKG (ref)
SIG-BMKG intensity thresholds grounded in Wald 1999, Worden 2012, Caprio 2015. Motivated the MMI-hybrid (<2 / 2–50 / ≥50 gal) boundary used by Stage 1 ensemble.
Threshold intensitas SIG-BMKG berlandaskan Wald 1999, Worden 2012, Caprio 2015. Memotivasi boundary MMI-hybrid (<2 / 2–50 / ≥50 gal) yang digunakan ensemble Stage 1.
Authors
Penulis
- Hanif Andi Nugraha — Department of Physics, Universitas Indonesia, Depok. Corresponding author. ORCID: 0009-0007-9975-1566. Departemen Fisika, Universitas Indonesia, Depok. Penulis koresponden. ORCID: 0009-0007-9975-1566.
- Dede Djuhana — Department of Physics, Universitas Indonesia, Depok / BMKG, Jakarta. ORCID: 0000-0002-2025-0782. Departemen Fisika, Universitas Indonesia, Depok / BMKG, Jakarta. ORCID: 0000-0002-2025-0782.
- Adhi Harmoko Saputro — Department of Physics, Universitas Indonesia, Depok. ORCID: 0000-0001-6651-0669. Departemen Fisika, Universitas Indonesia, Depok. ORCID: 0000-0001-6651-0669.
- Sigit Pramono — BMKG, Jakarta. ORCID: 0009-0000-5684-282X.
Manuscript target: IEEE Access (Q1). Reproducibility package: review_bundle.zip scheduled for Zenodo DOI archival. Updated 2026-04-24 with Task 15 final results.
Target manuskrip: IEEE Access (Q1). Paket reproducibility: review_bundle.zip dijadwalkan archival DOI Zenodo. Diperbarui 24-04-2026 dengan hasil akhir Task 15.