Upload SARIMAX(2, 1, 1)x(2, 0, 2, 12) — test RMSE 5.1561
Browse files- README.md +8 -11
- fig4_forecast.png +2 -2
- fig5_error_analysis.png +0 -0
- fig6_diagnostics.png +2 -2
- model_config.json +13 -12
- sarimax_model.pkl +2 -2
README.md
CHANGED
|
@@ -17,7 +17,7 @@ using historical water level data and meteorological variables.
|
|
| 17 |
|
| 18 |
| Parameter | Value |
|
| 19 |
|---|---|
|
| 20 |
-
| Architecture | SARIMAX(
|
| 21 |
| Seasonal period | 12 months |
|
| 22 |
| Target | Water level (m) |
|
| 23 |
| Exogenous variables | Temperature (°C), Precipitation (mm), Wind Speed (km/h) |
|
|
@@ -27,23 +27,20 @@ using historical water level data and meteorological variables.
|
|
| 27 |
|
| 28 |
## Hyperparameter Tuning
|
| 29 |
|
| 30 |
-
|
| 31 |
ranked by out-of-sample validation RMSE.
|
| 32 |
|
| 33 |
-
|
| 34 |
-
|---|---|
|
| 35 |
-
| Best validation RMSE | 3.9392 |
|
| 36 |
-
| Best AIC | 4624.6238 |
|
| 37 |
|
| 38 |
## Test Set Performance
|
| 39 |
|
| 40 |
| Metric | Value |
|
| 41 |
|---|---|
|
| 42 |
-
| RMSE | 5.
|
| 43 |
-
| MAE | 4.
|
| 44 |
-
| MAPE (%) | 6.
|
| 45 |
-
| R² | -0.
|
| 46 |
-
| NSE | -0.
|
| 47 |
|
| 48 |
> This model is a statistical baseline for benchmarking against
|
| 49 |
> deep learning approaches (LSTM, TCN).
|
|
|
|
| 17 |
|
| 18 |
| Parameter | Value |
|
| 19 |
|---|---|
|
| 20 |
+
| Architecture | SARIMAX(2, 1, 1)x(2, 0, 2, 12) |
|
| 21 |
| Seasonal period | 12 months |
|
| 22 |
| Target | Water level (m) |
|
| 23 |
| Exogenous variables | Temperature (°C), Precipitation (mm), Wind Speed (km/h) |
|
|
|
|
| 27 |
|
| 28 |
## Hyperparameter Tuning
|
| 29 |
|
| 30 |
+
Bayesian search : 50 trials | criterion: validation RMSE
|
| 31 |
ranked by out-of-sample validation RMSE.
|
| 32 |
|
| 33 |
+
|
|
|
|
|
|
|
|
|
|
| 34 |
|
| 35 |
## Test Set Performance
|
| 36 |
|
| 37 |
| Metric | Value |
|
| 38 |
|---|---|
|
| 39 |
+
| RMSE | 5.1561 |
|
| 40 |
+
| MAE | 4.1987 |
|
| 41 |
+
| MAPE (%) | 6.5872 |
|
| 42 |
+
| R² | -0.3838 |
|
| 43 |
+
| NSE | -0.3838 |
|
| 44 |
|
| 45 |
> This model is a statistical baseline for benchmarking against
|
| 46 |
> deep learning approaches (LSTM, TCN).
|
fig4_forecast.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
fig5_error_analysis.png
CHANGED
|
|
fig6_diagnostics.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
model_config.json
CHANGED
|
@@ -1,12 +1,12 @@
|
|
| 1 |
{
|
| 2 |
"model_type": "SARIMAX",
|
| 3 |
"order": [
|
|
|
|
| 4 |
1,
|
| 5 |
-
1
|
| 6 |
-
2
|
| 7 |
],
|
| 8 |
"seasonal_order": [
|
| 9 |
-
|
| 10 |
0,
|
| 11 |
2,
|
| 12 |
12
|
|
@@ -31,17 +31,18 @@
|
|
| 31 |
"n_obs": 96
|
| 32 |
},
|
| 33 |
"test_metrics": {
|
| 34 |
-
"RMSE": 5.
|
| 35 |
-
"MAE": 4.
|
| 36 |
-
"MAPE_pct": 6.
|
| 37 |
-
"R2": -0.
|
| 38 |
-
"NSE": -0.
|
| 39 |
},
|
| 40 |
-
"
|
| 41 |
-
"
|
|
|
|
| 42 |
"criterion": "val_RMSE",
|
| 43 |
-
"best_val_RMSE": 3.
|
| 44 |
-
"best_AIC":
|
| 45 |
},
|
| 46 |
"notes": "Basic SARIMAX \u2014 raw exog only, no lag features. Oracle exog assumption at forecast time."
|
| 47 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"model_type": "SARIMAX",
|
| 3 |
"order": [
|
| 4 |
+
2,
|
| 5 |
1,
|
| 6 |
+
1
|
|
|
|
| 7 |
],
|
| 8 |
"seasonal_order": [
|
| 9 |
+
2,
|
| 10 |
0,
|
| 11 |
2,
|
| 12 |
12
|
|
|
|
| 31 |
"n_obs": 96
|
| 32 |
},
|
| 33 |
"test_metrics": {
|
| 34 |
+
"RMSE": 5.1561,
|
| 35 |
+
"MAE": 4.1987,
|
| 36 |
+
"MAPE_pct": 6.5872,
|
| 37 |
+
"R2": -0.3838,
|
| 38 |
+
"NSE": -0.3838
|
| 39 |
},
|
| 40 |
+
"bayesian_optimization": {
|
| 41 |
+
"n_trials": 50,
|
| 42 |
+
"sampler": "Optuna TPE",
|
| 43 |
"criterion": "val_RMSE",
|
| 44 |
+
"best_val_RMSE": 3.9717,
|
| 45 |
+
"best_AIC": 4623.9462
|
| 46 |
},
|
| 47 |
"notes": "Basic SARIMAX \u2014 raw exog only, no lag features. Oracle exog assumption at forecast time."
|
| 48 |
}
|
sarimax_model.pkl
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c892e377360030ae904b9b068dc358a6f4a4ed9e690069efb0e747b62d91a6d9
|
| 3 |
+
size 121925408
|