Being a data-driven marketing platform, performance has always been at the core of our vision. At RevX, we use our proprietary machine learning model to predict the right bid price for the right user to deliver the most effective ads in real-time. Our team of data scientists strive to ensure that each campaign reaches its maximum potential and ROI goals. In this blog, we will give you a sneak-peak into what goes behind building a sophisticated prediction model and the challenges that come its way.
We model the problem of predicting an ad click as a binary classification problem with two possible outcomes – ad is clicked or not. The prediction model typically returns an estimation of the outcome as a probability between 0 and 1 given the context in which the ad will be served. Starting with some initial seed weights for each feature, the model is trained using an optimization algorithm like gradient descent and feature weights are continuously updated based on the gradient of a cost function like log loss. The training iterations continue until the model returns the lowest possible value of the log loss function. Traditional optimization algorithms use the entire batch of training samples to adjust the features weights in every iteration and are often termed as batch learning algorithms.