Pan NA

Monson, Tesla A. & Hlusko, David W. Armitage and Leslea J., 2018, Using machine learning to classify extant apes and interpret the dental morphology of the chimpanzee-human last common ancestor, PaleoBios 35, pp. 1-20 : 10-12

Pan NA Pan

Homo NA Pan

Homo NA Pan

Homo Homo Pan

Homo NA Pan length and dental area are majority Homo .

Uniquely among the Miocene fossil sample, Micropithecus clarki Fleagle and Simons, 1978 is classified as Pan with 100% agreement using dental length, area, and the MMC and PMM ratios. On the opposite end of the spectrum, Ankarapithecus meteai Ozansoy, 1957 is classified as Gorilla with 100% agreement using dental length, area, and the MMC and PMM ratios. Khoratpithecus piriyai Chaimanee et al., 2004 is also classified as Gorilla with 100% agreement using dental length, and the MMC and PMM phenotypes (dental areas are not available for this taxon).

Like many of the fossil specimens, Ardipithecus is classified as Pan using dental length, and as Gorilla using the MMC and PMM ratios. In contrast, Australopithecus robustus Broom, 1938 is almost exclusively classified as Gorilla by the machine learning LDA model ( Fig. 1). The other Australopithecus specimens have less agreement between data sets. Many of the Au. afarensis Johanson and White, 1979 specimens are classified exclusively as Gorilla using all three data types, while some of them are classified as Homo using dental length and dental area, and as Gorilla using the MMC and PMM phenotypes. All three of the Australopithecus bosei Leakey, 1959 specimens are exclusively classified as Gorilla .

Interestingly, there is good agreement on the classification of Homo habilis (sensu lato) Leakey et al., 1964 as Gorilla using all of the phenotypes except for OH-13 which is classified as Homo using dental length and dental area. In contrast, Homo antecessor Bermúdez de Castro et al., 1997 is classified as Pan with 100% agreement using dental length, area, and the MMC and PMM ratios. There is more variation in the other species of Homo although many of the individuals are classified as Pan using dental length and area. Homo erectus Mayr, 1951 is largely classified as Homo using dental length and as Pan with the MMC and PMM ratios. Homo heidelbergensis Schoetensack, 1908 is jointly classified as Pan and Homo using dental length and area, but the sample is classified as Pan , Gorilla , or Pongo using the MMC and PMM phenotypes. Homo neanderthalensis King, 1864 is almost exclusively classified as Pan using dental length, but is jointly classified as Pan and Homo using MMC and PMM. Overall, many of the H. erectus , H. heidelbergensis , and H. neanderthalensis specimens are classified as Homo using dental length, emphasizing the overall similarity of tooth size between these taxa and modern humans. However, the dental proportions of fossil Homo fall at the intersection of modern apes ( Homo , Gorilla , and Pan ) and tend to be more variably classified by the machine learning algorithm. Classifications of each specimen using dental length, dental area, and the MMC and PMM ratios are fully detailed in Table 5.

Because machine learning is not static, multiple iterations of the method will result in slight changes of classification. The training sample also plays an important role in the method, and a larger, or different, extant sample would likely have some impact on the classification analysis of the fossil taxa. As we note here, the phenotypes used in the method also dramatically influence the results of the classification.