Ad-Click Prediction

November - December 2021

​

Authors: Steven Sun, Jarvis Huang, Michelle Wan,
               Zed Kamurase Jean Bosco

​

Advisors: Prof. George Easton

​

Project Overview:​

• Examined over 32MM labeled training dataset and 20MM unlabeled test dataset with 23 attributes related to online advertisement impressions

• Developed embedded layer architecture in TensorFlow to transform categorical features into multidimensional vectors quantifying cross-category correlation

• Implemented neural network architecture and experimented model behavior over different hyperparameters (mini-batch gradient descent, L2 regularization, etc.)

• Reached log loss of 0.398 on test data in terms of model performance

​