I believe that hands-on experience is necessary to internalize basic principles. Thus when I teach data science, I like to give each student team an interesting but messy forecasting challenge, and demand that they build and evaluate a predictive model for the task.
These forecasting challenges are associated with events where the students must make testable predictions. They start from scratch: finding the relevant data sets, building their evaluation environments, and devising their model. Learn more Data science online training
Finally, I make them watch the event as it unfolds, to witness the vindication or collapse of their prediction. As an experiment, we documented the evolution of each group’s project on video in Fall 2014. Professionally edited, this became The Quant Shop, a television-like data science series for a general audience.
• Finding Miss Universe – The annual Miss Universe competition aspires to identify the most beautiful woman in the world. Can computational models predict who will win a beauty contest? Is beauty just subjective, or can algorithms tell who is the fairest one of all?
• Modeling the Movies – The business of movie-making involves a lot of high-stakes data analysis. Can we build models to predict which film will gross the most on Christmas day? How about identifying which actors will receive awards for their performance?
• Winning the Baby Pool – Birth weight is an important factor in assessing the health of a newborn child. But how accurately can we predict junior’s weight before the actual birth? How can data clarify environmental risks to developing pregnancies?
• The Art of the Auction – The world’s most valuable artworks sells at auctions to the highest bidder. But can we predict how many millions a particular J.W. Turner painting will sell for? Can computers develop an artistic sense of what’s worth buying?
• White Christmas – Weather forecasting is perhaps the most familiar domain of predictive modeling. Short-term forecasts are generally accurate, but what about longer-term predictions? What places will wake up to a snowy Christmas this year? And can you tell me one month in advance?
• Predicting the Playoffs – Sports events have winners and losers, and bookies are happy to take your bets on the outcome of any match. How well can statistics help predict which football team will win the Super Bowl? Can Google’s PageRank algorithm pick the winners on the field as accurately as it does on the web?
• The Ghoul Pool – Death comes to all men, but when? Can we apply actuarial models to celebrities, to decide who will be the next to die? Similar analysis underlies the workings of the life insurance industry, where accurate predictions of lifespan are necessary to set premiums that are both sustainable and affordable. Data science course
• Playing the Market – Hedge fund quants get rich when guessing right about tomorrow’s prices, and poor when wrong. How accurately can we predict future prices of gold and oil using histories of price data? What other information goes into building a successful price model?
I encourage you to watch some episodes of The Quant Shop in tandem with reading this book. We try to make it fun, although I am sure you will find plenty of things to cringe at.
Each show runs for thirty minutes and maybe will inspire you to tackle a prediction challenge of your own. These programs will certainly give you more insight into these eight specific challenges. I will use these projects throughout this book to illustrate important lessons in how to do data science, both as positive and negative examples.
These projects provide a laboratory to see how intelligent but inexperienced people not wildly unlike yourself thought about a data science problem, and what happened when they did.
