Evergreen Ballroom Prefunction
A Random Forest Classification of Prehistoric Stone Tools in South Texas (306744)
Juan Luis Gonzalez, University of Texas Rio Grande ValleyJames Hinthorne, University of Texas Rio Grande Valley
*Brandi Michelle Reger, University of Texas Rio Grande Valley
Hansapani S Rodrigo, University of Texas Rio Grande Valley
Keywords: geochemical cluster analysis, XRF, predictive classification model, random forest
The identification of lithic materials plays an important role in archeology, but visual identification of these lithic materials is not reliable. In this project, we report efficient clustering approaches and a development classification model to identify lithic materials using their chemical properties. We have utilized over 1000 XRF scans of artifacts and a local raw material (ESC). The clustering algorithms Partition Around Medoids (PAM) and Fuzzy C-Means (FCM) were found to best preserve the classification of the known material (ESC). This process generated 5 clusters, which were used to train a random forest (RF) model with a precision rate of 91%. A RF model trained on the results of visual classification alone identified samples in 3 clusters with a precision of 82%. This model identified several classification errors made during the visual analysis stage as well as various incorrect assumptions made during the interpretation of the clustering results. This has proven to be a useful approach for improving the accuracy of lithic material analysis over visual classification alone.