https://www.dau.edu
TABLE 3. PROGRAMS UTILIZED IN THE STUDY (CONTINUED)
UH-1H Utility Military Helicopter Composite Main Rotor Blade Development AH-64E Apache Attack Helicopter (Formerly AB3) Flight Test Build
B-1B Heavy Bomber - Conventional Mission Upgrade Program
F136 Advanced Turbofan Engine - Prototype
F/A-18E/F Multirole Fighter Aircraft - Distributed Targeting Processor
GE38-1B Gas Turbine Engine
USMC H-1 Helicopter Upgrade for AH-1Z (4BW) TF34-GE-100 Turbofan Engine
E-2D Advanced Hawkeye Tactical Airborne Early Warning Aircraft Home Station Instrumentation Training System (HITS) AN/APG-79 Active Electronically Scanned Array Radar - Prototype Program Definition & Risk Reduction (PDRR)
Model 214A (UH-1 Utility Helicopter) - Foreign Military Sales Iran Full Scale Development
M-898 - Sense and Destroy Armor (SADARM) 'Smart' Submunition
E-2C Tactical Airborne Early Warning Aircraft - Mission Computer Upgrade Program
The distributions presented in the ToLR consist of the lognormal, normal, Weibull, triangular, uniform, and beta. For our purposes, we utilize the lognormal, Weibull, Rayleigh (a specialized Weibull distribution), beta, and triangular. We adopt the method of maximum likelihood to estimate the parameters of these distributions. Maximum likelihood is a method of curve fitting to a function with unknown parameters whereby the data are the most probable to occur when randomly sampled from the given distribution with the resulting parameters (Myung, 2003). To compare which distributions best fit the cost growths for each element or WBS, the Akaike Information Criterion (AIC) for each fit is compared across our five distributions. The best fit distribution is the distribution with the lowest AIC, which is an estimator of prediction error and thereby relative quality of statistical models for a given set of data (Stocia & Selen, 2004). AIC considers both the quality of the model and the number of parameters required by the distribution, penalizing the AIC score for distributions that require a greater number of parameters. In this way, AIC helps to avoid selecting overfitted distributions. The best overall distribution is determined by the lowest mean rank for each element or WBS. We next display our results. Results Primarily using the Python libraries of Pandas, SciPy, NumPy, and the JMP 15 Pro software, we produce the distributional fitting outputs presented in Tables 4 and 5. Table 4 lists the AIC for the lognormal,
41
Defense ARJ , Spring 2025, Vol. 32 No. 1
Made with FlippingBook - Online Brochure Maker