| Abstract Detail
Biodiversity Informatics & Herbarium Digitization Tyrrell, Christopher [1]. Next generation identification keys: A method for implementing continuous characters and estimating the probability of an ID. Species identification is vital to many disciplines and digital technology has improved the tools used for identifying species. Most digital identification keys, even multi-access character-matrix based keys rely on discrete character states. The direct use of characters with continuous states has yet to be fully realized. To achieve full use of continuous characters for identification, I propose using a naive Bayesian classifier (NB). The NB provides both a taxonomic annotation and a posterior probability (degree of belief) in the name assignment. The classifier works by formulating a species class as a multivariate model of morphological characters. Each character (continuous or discrete) has a probability function defined for it and the NB algorithm uses these to calculate a degree of belief for all candidate species in the key. The optimal annotation is that species class with the maximum a-posteriori probability (MAP). To gauge the relative evidence for the identification with the MAP, I introduce a log ratio of probabilities that estimates the strength of the species assignment. I apply the proposed method to two example identification keys: native vs. invasive Myriophyllum in North America and vegetative Rhipidocladum bamboos in Mexico. These examples are chosen because they both contain taxa that are difficult to differentiate using discrete character states. In each instance, the novel method provides a probability and estimate of the strength of the probability to qualify the name assigned. Similar to statistical identification methods like image recognition, the NB uses machine learning techniques to advance our ability to use digital technology for improved, interactive taxonomic identifications. Unlike image-based keys, however, this method is directly interpretable and takes advantage of the predictive information inherent in the morphological character data of monographs and species descriptions. Log in to add this item to your schedule
1 - Milwaukee Public Museum, Botany Department, 800 W Wells St, Milwaukee Public Museum, Milwaukee, WI, 53233, United States
Keywords: key morphology taxonomic Bayesian identification.
Presentation Type: Oral Paper Session: BIHD1, Biodiversity Informatics & Herbarium Digitization Location: Tucson I/Starr Pass Date: Monday, July 29th, 2019 Time: 1:30 PM Number: BIHD1001 Abstract ID:89 Candidate for Awards:None |