Estimating dollar-value outcomes of workman`s compensation claims using radial basis function networks

Conference · Sun Dec 31 04:00:00 EST 1995

OSTI ID:377066

Hancock, Jr, M F ^[1]

---

1. Rollins College, Winter Park, FL (United States)

The National Council on Compensation Insurance (NCCI) maintains a national data base of outcomes of workers` compensation claims. We consider whether a radial basis function network can predict the total dollar value of a claim based upon medical and demographic indicators (MDI`s). This work used data from 12,130 workers` compensation claims collected over a period of four years from the state of New Mexico. Two problems were addressed: (1) How well can the total incurred medical expense for all claims be predicted from available MDI`s? For individual claims? (2) How well can the duration of disability be predicted from available MDI`s? The available features intuitively correlated with total medical cost were selected, including type of injury, part of body injured, person`s age at time of injury, gender, marital status, etc. These features were statistically standardized and sorted by correlation with outcome valuation. Principal component analysis was applied. A radial basis function neural network was applied to the feature sets in both supervised and unsupervised training modes. For sets used in training, individual case valuations could consistently be predicted to within $1000 over 98% of the time. For these sets, it was possible to predict total medical expense for the training sets themselves to within 10%. When applied as blind tests against sets which were NOT part of the training data, the prediction was within 15% on the whole sets. Results on individual cases were very poor in only 30% of the cases were the predictions for the training sets within $1000 of their actual valuations. Single-factor analysis suggested that the presence of an attorney strongly decorrelated the data. A simple stratification was performed to remove cases involving attorneys and contested claims, and the procedures above repeated. Preliminary results based upon the very limited effort applied indicate that NCCI data support population estimates, but not single-point estimates.

• Research Organization: Pacific Northwest Lab., Richland, WA (United States)
• OSTI ID: 377066
• Report Number(s): PNL-SA--26375; CONF-9503142--; ON: DE96009360
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
COST ESTIMATION
NEURAL NETWORKS
WORKMENS COMPENSATION