Discipline: Mathematics and Statistics
Subcategory: Mathematics and Statistics
Session: 1
Room: Tyler
Marqus Parker - North Carolina Agricultural and Technical State University
Handwriting is one of the most common means of communication in daily life. Digital recognition of both online and offline handwriting is critical in many applications such as bank check readability, signature verification, and touch-screen writing recognition. Although researchers are interested in detecting hidden characteristics in handwriting such as gender, handedness, age, and ethnicity, this research field remains unfledged. The object of this study is to improve the detection of hidden characteristics in handwriting using real data with a hybrid method of statistical learning algorithms and resampling techniques. For this study, we collect alpha-numerical character data from 300 college students. The collected data are converted digital format and processed for further analysis. We generate three different training data sets, original, oversampled, and undersampled data. Selected learning algorithms including a family of K-Nearest Neighbors and artificial neural network are applied to those training data sets to construct predictive algorithms. The prediction performance of the algorithms constructed is compared in the testing data. Our results will show which statistical learning algorithm achieves the highest accuracy and how resampling methods contribute to the improvement of algorithms for detecting selected characteristics. Our study is meaningful because we propose a hybrid method of statistical learning and resampling which is applied to our purposefully collected data to achieve a more accurate detection of unobserved traits in handwriting.
Funder Acknowledgement(s): NSF ACE DSA Project and Mathematics Department at North Carolina A&T State University
Faculty Advisor: Dr. Seongtae 'Ty' Kim, skim@ncat.edu
Role: I conducted every part of this research experiment.