A self adaptive cognitive deep learning framework for classifying graphology features to Big five personality traits

Abstract

Graphology is a technique for study and analysis of individual personality from his/her handwriting style. Most of the existing graphology-based solutions for personality detection recognize nonstandard application dependent personalities. Even the very few big five personality recognition approaches have limited accuracy and lacks adaptivity to new handwriting styles. Towards these problems, a novel self-adaptive cognitive learning framework based on deep learning convolutional neural network (CNN) features is proposed to classify the handwritten document to big five personality traits. This framework correlates the various document level and character level graphology features to big five personality traits to recognize features with strong correlation to various big five personality traits and uses these features to classify the personality. To enhance the deep learning feature learning ability an enhanced convolution kernel is proposed for the CNN. Through testing with various handwritten documents, the proposed solution is found to provide 2.18% higher accuracy and 5% lower false positives compared to existing works on big five personality classification.