Implementation of Hashing Algorithms in Stream Mining

Abstract

Artificial intelligence in food logistics has shown potential to substitute the First In First Out approach with much more promising First Expiring First Out approach. However, the shelf life prediction might be hindered by the amount of storage needed and by the amount of processing time elapsed between an event and appropriate system response. We use the data structure Count-Min (CM) sketch for approximating frequencies of critical ambient parameters (CAPs) in the cold-chain of peaches and nectarines. The CM sketch supports fast updates, but it also supports a trade-off between compression and error. This is important in the cold-chain setting because of the limited storage space in sensor nodes, but also because transmitting each sensed value increases energy consumption and introduces delays. In order to assess the trade-off between memory consumption and information loss, the actual values of CAPs with their estimates in various parameter settings are presented graphically. The estimates are obtained by querying the point estimator with the original dataset values. Errors are computed as differences between actual values and their estimates, and scaled by the number of instances.