An Application of Data Mining Algorithms for Shipbuilding Cost Estimation
AP02A_Kaluzny SCEA Conference Paper (revision)
AP02_Kaluzny ISPA SCEA 2011 slides (revision)
The North Atlantic Treaty Organization (NATO) Research and Technology Organization (RTO) Systems Analysis and Studies (SAS) 076 Panel (NATO Independent Cost Estimating and its Role in Capability Portfolio Analysis) is a working panel generating independent cost estimates for NATO systems with the aim of standardizing how NATO countries conduct cost estimation. One of the systems analyzed by the SAS-076 Panel in an ex post exercise was Her Netherlands Majesty’s Ship (HNLMS) Rotterdam Landing Platform Dock (LPD), an amphibious transport dock ship that was launched in 1997.
Technical and cost data for ships similar to the HNLMS Rotterdam LPD was gathered. A database of 59 ships in 18 classes from 7 nations was compiled, spanning years (commissioned) from 1954 to 2010. For each ship, over a hundred descriptive and technical ship attributes were obtained, encompassing dimensions, performance, power generation, lift capacity, armament & countermeasures, sensors, combat & weapon control systems, etc. Ship development and production cost data was expressed in various currencies and a normalization procedure was required.
Two independent cost estimating methods were used to generate cost estimates:
1) Parametric approach: Traditional ship building cost estimating relationships (CERs) are often mathematically simple (e.g. a simple ratio) or involve linear regression analysis (of historical systems or subsystems) on a single parameter (weight, length, density, etc.). This is often insufficient – other cost driving factors must be incorporated to develop estimates of sufficient quality at the preliminary design phase. Furthermore, the relationship between the parameter(s) and cost may not be best expressed in linear form. While the field of regression analysis offers a multitude of alternative approaches, linear regression is the most popular and easiest to understand. A parametric approach for ship cost estimation that incorporates a multitude of cost driving factors, while remaining a top-down approach applicable in early design phases of the procurement cycle was used. The M5 Model Tree Algorithm combines features of decision trees with linear regression models to both classify similar ships and build piece-wise multivariate linear regression models.
2) Costing by Analogy: Cost estimation by analogy is typically accomplished by forecasting the cost of a new system based on the historical cost of similar or analogous system. There must be a reasonable correlation between the new and historical system. The cost of the historical system is adjusted by undertaking a technical evaluation of the differences between the systems, deducting the cost of components that are not comparable to the new design and adding estimated costs of the new components. Subject matter experts are required to make a subjective evaluation of the differences between the new system of interest and the historical system. Subjectively chosen complexity factors are often used to adjust the analogous system’s cost to produce an estimate. The credibility of the estimate for the new system may be undermined if the adjustment factors are not substantiated – this is a key disadvantage of the traditional analogy method. Hierarchical cluster analysis was used for a novel cost estimation by analogy approach void of the subjectivity inherent (of the traditional approach) in quantifying the cost of the technical and other differences between the historical system and the new system. The approach also considers multiple analogous systems rather than just one.
Cost estimates for the HNLMS Rotterdam LPD were generated using to two approaches. Once the estimates were finalized, the Royal Netherlands Navy disclosed the actual development and production cost of the HNLMS Rotterdam for comparison. In this talk we present the theory and application of the two cost estimation methods, and discuss the ex post analysis results.
Bohdan L. Kaluzny
Defence Research & Development Canada, Centre for Operational Research & Analysis
Dr. Bohdan Kaluzny is a defence scientist from Defence Research & Development Canada, Centre for Operational Research & Analysis. He currently heads a team of three scientists providing support to the Assistant Deputy Minister (National Defence) of Materiel, specifically to the major Crown acquisition projects. He has six years experience in military and security operational research. Dr. Kaluzny has a doctorate in computational science from McGill University and has published articles in the fields of computational geometry, combinatorial optimization, and multi-criteria decision analysis. He is a member of NATO Research & Technology Systems Analysis and Studies 076 Panel Task Group: NATO Independent Cost Estimating and its Role in Capability Portfolio Analysis.