As we enter into the countless debates on whether or not the development of “driverless” cars is great for all those people who make money by driving Uber or Lyft, I’ve been thinking on a much more fundamental question—Are these discussions even relevant for the drivers? Not really, at least not in the next 10-15 years. So what is more relevant to them, today?
Here are two of the most important questions which, I am sure, most of the Uber/Lyft Drivers would love to know the answers to:
How much will I earn by the end of the week?
Is there a way to crack the code to ‘Drive less and make more money’?
So let’s switch roles to address these questions. Shall we? Let’s say a data scientist decides to start driving an Uber. I want to present to you a case highlighting how ‘he/she’ would think. Let’s assume that, I (data scientist) decided to drive and was trying to figure out the answers to the above two questions. (And Yeah, since I work for SAP, I also have access to our machine-learning platform called SAP BusinessObjects Predictive Analytics to make me a Smart Uber Driver – Advantage Surya!)
Like any other analysis, this analysis also needs and starts with data. I live in the city of Chicago and, assuming that I drove Uber for an entire year, saw all different seasons (Especially the winters!), and potentially drove to all parts of the cities, I would have data related to all the above pointers for the last 362 days, along with my earnings, weather, holidays, weekends, GPS information, and so on. Here is a visual representation of Chicago on a map. This is just to give you an idea of how different suburbs are spread out and how far the airport is from Downtown Chicago.
So, now that we understand the data and the map as well as the routes, let’s get straight to the questions.
Question 1 – Given my earnings data for the last 1 year, how much money will I make ‘daily’ for the next 5 days?
Ha! This is a forecasting problem, isn’t it? Well, to solve this problem, I would need the historical data of my earnings. Here is how I see my data in Uber App—a clean, nice interface that gives me my daily earnings for the entire year.
So, I downloaded my own data and used SAP BusinessObjects Predictive Analytics’ time-series forecasting wizard to actually build a forecasting model in a very short amount of time.
So, I downloaded my own data and used SAP BusinessObjects Predictive Analytics’ time-series forecasting wizard to actually build a forecasting model in a very short amount of time. Here, I forecasted only for 5 days and this is how my model performed. We had data from Oct 1st 2015 till Oct 1st 2016 (362 records). We only took 7 seconds to build a model to predict my earnings for next 5 days.
SAPBusinessObjects Predictive Analytics was able to not only forecast the earnings, but also determine if there was a trend and seasonality hidden in my data.
What about the Quality of the Model, and the Trends Revealed?
So, how do I know if my model is good ? Introducing MAPE—this quality indicator for the forecasting model is the mean of MAPE values observed over all the training horizon. A value of zero indicates a perfect model while values above 1 indicate bad quality models. Here we’ve got a model with a 3.3% error rate, that’s pretty good!
What about the Trend and Seasonality ? Our Model identified a Polynomial trend, which is a cycle that repeats every 29 days. (Maybe that reflects the PayDay!?) So, we can hypothesize that people are going out a lot on the day when they get their salary!
Let’s look at the forecasting output now:
You can clearly see a pattern in my data set. Obviously, I started to make much more money than I made earlier, but you can also see a constant peak during the weekends and a higher spike during the first weekend of every month.
This analysis was very interesting because I could see a linear positive trend which is shown in the graph below.
Hmmm… Interesting… So now I have the answer to my first question which was, "How much money am I going to make daily for the next 5 days?"
In Part 2 of this blog, I'll move on to the second question, "Is There a Way to Crack the Code to ‘‘Drive Less and Make More?’" Stay tuned!