Models

Why "plain" DEA isn't enough

The Hurst parameter and fractal dimension are ratio-form variables. Classical DEA models (CCR/BCC) violate convexity, proportionality and monotonicity when evaluating this kind of variable, producing incorrect efficiency frontiers. That's why each stage of the research proposed a multiplicative variant — with a logarithmic transformation — suited to this problem.

Decision variables

The same three variables, across every model

Input · minimize

Fractal Dimension (D)

Measures traffic smoothness/irregularity. Ranges from 1 to 2: the closer to 1, the more stable; the closer to 2, the more irregular and bursty.

Output · maximize

Hurst Parameter (H)

Measures the series' time memory. H > 0.5 indicates long-range dependence (the pattern tends to repeat in the future) — only series with H > 0.5 are useful for prediction.

Output · maximize

Average TCP Throughput

Average bandwidth measured by iperf/iperf3 across each configuration's (DMU) time series.

Why WMDEA outperforms classical DEA

Three concrete advantages over CCR and BCC

See the practical result of these advantages in the data →

Model family

From classical DEA to Super-Cobb-Douglas

Qualitative comparison

Advantages and limitations, side by side

ModelOrientation / mechanicsAdvantageLimitation