I have a certifaction in machine learning from UW through Coursera. In my coursework, I built classification, regression, and clustering algorithms from scratch using data structures from GraphLab Create and Scikit-Learn. I used these algorithms to accurately predict house prices, sort Wikipedia pages, and determine the sentiment of Amazon reviews.
For work at my affiliate startup we needed an algorithm that could predict the trajectory of financial growth. Not knowing how the growth could vary over time, I devised a system where more recent data points could be given progressively more predictive power over older, less relevant data points.
To accomplish this, I used a special case of simple linear regression that incorporates a weight vector. Here is a link to my post explaining the mathematical process I used to derive this formula.