Optimizing Mobile Advertising with Reinforcement Learning and Deep Neural Networks

Mobile advertising is also closely associated with contemporary digital marketing channels, where the projective targeting is critical in regard to businesses’ potential performance. This research makes use of a combination of machine learning and reinforcement learning (RL) to determine the best place to post an ad considering the users’ demographic and behavior. Deepening the neural networks (DNNs) in this regard, the method is expected to advance the efficiency of mobile advertising while delivering more relevant content to the target user. Purchasing tendency predictions in the report are based on Python-based models through a dataset from Kaggle that has incorporated user demographics (Gender, Age, Estimated Salary) and purchasing patterns. The missing values are first handled using the different techniques then normalization of the numerical data and one hot coding of the features such as the Device Type and the Location. Automatic feature are extracted with the help of DNNs which obviate the process of feature extraction. This approach proposed yielded an accuracy level of 98% a higher level as compared to the CNN and LSTM models. For decision-making reinforcement learning is used with the help of Deep Q-Network (DQN) and Policy gradient. DQNs predict Q-values for the actions (ad displays) and choose the best action, while Policy Gradient methods directly learn the policy by modifying the probabilities related to the actions based on the users’ behavior. The training techniques used on both models include backpropagation and stochastic gradient descent, which enables the RL agent in the continuous implementation of better ad placement strategies. Here they used DNN to train the features and RL to optimize the training and then improve the on average click through rates and conversion rates which provides a better strategy towards advertising.