For automatic, rapid, accurate and objective classification of fish freshness under cold storage an electronic nose using seven metal dioxide gas sensors was developed to detect fish volatiles. Total viable count and Total volatile base nitrogen analyses were conducted simultaneously to indicate fish quality status. By sampling fish headspace, patterns were obtained during 15 storage days. 35 appropriate odor parameters were seleected from each test. Principle component analysis was applied to reduce the 35-dimensional vectors to 5-dimensional vectors and clustered samples into fresh, semi fresh and spoiled. With 5-dimensional vectors as input, multilayer perceptron neural network modeled fish spoilage based on these three classes with 96.87 percent correct rate of test data. We found that the newly introduced hyper disk models maximum margin optimum classifier yielded 100 percent correct rate that could be successfully applied in industry for the diagnosis of fish spoilage.

Keywords