Gender Classification with the First Name of the People as a Feature using Character Based 1D Convolutional Neural Network

Abstract

The gender identification based on the name of a person is a traditional problem which many researchers are trying to address using different models and techniques. Convolutional Neural Network (CNN) is a powerful deep learning technique which plays a huge role in image classification and signal processing domains. The aim of this thesis is to introduce a character level onedimensional (1D) Convolutional Neural Network model for the classification of the gender based on the name. Existing studies of the text classification using CNN are focused on sentence classification tasks with the help of two-dimensional (2D) Convolutional layers. In this study, different experiments have been administered by adjusting the model, activation functions and the dataset, in order to find an optimum model. The final model consists of 3 parallel Convolutional layers followed by max-pooling layers and 2 fully connected layers with a dropout layer in the middle. The model has been trained and tested with openly available US Census name, gender dataset. The occupied average validation accuracy is around 0.89