Handwriting analysis for personality trait features identification using CNN

Abstract

Handwriting analysis is an approach to get information through the handwriting. It extremely useful information, for instance in personality traits identification. The information came from the feature extracted from the handwriting. This feature can be size, slantness, pressure, and so forth. In this research, handwriting analysis is through the AND dataset that provide handwriting dataset along with feature label while most public dataset has nothing with it. By using the Coonvolutional Neural Networks (CNN) potentiality in capturing and recognizing global features, there are 15 models had built in this research in accordance with each feature and divided into three group by its number of types. After built a simple CNN architecture by only conduct two convolution layer, overall result show fair enough performance where the highest rate of accuracy is 80.88%. Furthermore, there are three best features had recognized, which is "entry stroke 'A'", "size", and "slantness", where the last two is naturally global features. However, the fact that handwriting image data cannot be oversampled which can lead to the bias result, than the imbalance data becomes a problem in this research that reduced the model performance.