Probabilistic models of food microbial safety and nutritional quality

Microbial and compositional analyses of processed foods, raw materials, ingredients and discharge streams are an important part of safety and quality assurance in the food industry. Law mandates that their results ought to be kept for a specified time. The records are usually in the form of charts or tabulated values presented or listed in a sequential order. Such records frequently appear as randomly fluctuating time series. Usually, the fluctuations are within a range considered safe or acceptable, and as long as the entries are all within this range the nature of the fluctuations is of little interest. Occasionally, there are exceptional entries; a high microbial count or a particularly low concentration of an important nutrient, for example. In most cases they can be traced to a known cause, equipment failure and human error are the most common. But there can be odd entries, which may have serious safety or quality implications that have no apparent reason. These occur because of the random coincidence of several factors, some unknown or undocumented. Usually they cancel one another, but not exactly, and hence the fluctuations. The probability that factors that tend to spoil the product, or lower its nutritive value, will act in unison can be estimated from the distribution of past events, provided that the entries are independent and their series stationary. The concept is demonstrated with industrial microbial and other records having symmetric and asymmetric distributions. It was tested by comparing the estimated frequencies of large or small values with those actually observed in fresh data. (C) 2002 Elsevier Science Ltd. All rights reserved.