While some changes have been made to protect intellectual property, the machine learning model that you're going to build today is based on an algorithm that is in use today by top investment banks and has been used to advise clients like Macy's, McDonald's, Nordstrom, the paint company PPG, the jewelry retailer Tiffany and Co, and some of the largest companies on earth. This is a real algorithm that is adding real value in investment banking advisory work today. Here's the situation. You're going to play the role of an investment banking analyst on a team that advises large corporations on raising capital. Your clients who are the treasurers and the CFOs of these companies often ask your team's advice about how much liquidity they should maintain and how they stack up against their peer companies. Liquidity is essentially corporate spending power. Your personal equipment would be the cash in your bank account plus your credit card limit. For a large corporation, maintaining excess liquidity is expensive, but maintaining too little is risky because if liquidity hits zero, the company goes bankrupt. This is exactly the kind of optimization problem where machine learning thrives. Historically, your team has provided advice to clients with a simple linear regression based on data from other public companies. However, as you can see below, this analysis is not very effective with an R squared of only 0.48. That means that this model only explains 48% of the variation in the variable that you're trying to predict. Your objective is to construct a machine learning regression algorithm that outperforms this existing analysis. You might have thought machine learning was only used for making predictions and wonder why it would be used to advise a corporation on what their liquidity should be today. If you're thinking that you're right, machine learning is used for predictive analysis. Here's the idea behind this liquidity regressor analysis. On average, public companies maintain a healthy amount of liquidity that prevents them from going bankrupt without excessive expenses.

Therefore, if we train a machine learning regression algorithm on data from these public companies, the algorithm will learn the collective wisdom of the market.

Once trained, you can give the algorithm your client's inputs, and the algorithm can estimate your client's ideal amount of liquidity. According to that collective wisdom of the market, the final amount of liquidity your client's treasurer or CFO chooses to maintain may be adjusted based on qualitative factors or non-public information. However, this machine learning analysis is the ideal quantitative tool to advise clients on how they compare to other public companies. To complete your machine learning algorithm, you're going to use a new package. Sckiti-learn the most popular Python machine learning package. Among other things, Sckiti-learn includes tools for machine learning, regression and model selection. You'll put these tools to work later in this lesson.