Comparative Analysis of Deep Learning Models for Real-World ISP Network Traffic Forecasting

These configs were used in the paper "Koumar, J., Smoleň, T., Jeřábek, K. and Čejka, T., 2025. Comparative Analysis of Deep Learning Models for Real-World ISP Network Traffic Forecasting. arXiv preprint arXiv:2503.17410.".

Benchmark hash	Dataset	Aggregation	Source
2439d12c2292	CESNET-TimeSeries24	1 HOUR	INSTITUTIONS
63882fe052f8	CESNET-TimeSeries24	1 HOUR	INSTITUTIONS
5a79f6cd3506	CESNET-TimeSeries24	1 HOUR	INSTITUTIONS
5a7471b2c70b	CESNET-TimeSeries24	1 HOUR	INSTITUTIONS
0f4fbc0419ce	CESNET-TimeSeries24	1 HOUR	INSTITUTIONS
d166d3b19a87	CESNET-TimeSeries24	1 HOUR	INSTITUTION_SUBNETS
2112383abab7	CESNET-TimeSeries24	1 HOUR	INSTITUTION_SUBNETS
5d9e6e63cbf0	CESNET-TimeSeries24	1 HOUR	INSTITUTION_SUBNETS
d82139cd671f	CESNET-TimeSeries24	1 HOUR	INSTITUTION_SUBNETS
d03ba8db5892	CESNET-TimeSeries24	1 HOUR	INSTITUTION_SUBNETS
2e92831cb502	CESNET-TimeSeries24	1 HOUR	IP_ADDRESSES_SAMPLE
702e58166879	CESNET-TimeSeries24	1 HOUR	IP_ADDRESSES_SAMPLE
394603854070	CESNET-TimeSeries24	1 HOUR	IP_ADDRESSES_SAMPLE
420d4303f949	CESNET-TimeSeries24	1 HOUR	IP_ADDRESSES_SAMPLE
e2c2148a178c	CESNET-TimeSeries24	1 HOUR	IP_ADDRESSES_SAMPLE

Example of usage of this related works configs:

from cesnet_tszoo.benchmarks import load_benchmark
from cesnet_tszoo.utils.enums import FillerType, ScalerType

benchmark = load_benchmark("2439d12c2292", "../")
dataset = benchmark.get_initialized_dataset()

# Get related results
related_results = benchmark.get_related_results()
print(related_results)

# Process with your own defined model
results = []
for ts_id in tqdm.tqdm(dataset.get_data_about_set(about='train')['ts_ids']):
    model = SimpleLSTM().to(device)
    model.fit(
        dataset.get_train_dataloader(ts_id), 
        dataset.get_val_dataloader(ts_id), 
        n_epochs=5, 
        device=device,
    )
    y_pred, y_true = model.predict(
        dataset.get_test_dataloader(ts_id), 
        device=device,
    )

    rmse = mean_squared_error(y_true, y_pred)
    results.append(rmse)

_mean = round(np.mean(results), 3)
_std = round(np.std(results), 3)
print(f"Mean RMSE: {_mean}")
print(f"Std RMSE: {_std}") 

# Compare with related works results
better_works = related_results[related_results['Avg. RMSE'] < _mean]
worse_works = related_results[related_results['Avg. RMSE'] <= _mean]
print(better_works)
print(worse_works)