Long-term Cycling Performance Analysis Using Strava Data

Posted on Mar 10, 2021

Strava is the most popular online platform for cylists to upload and analyze their data. There is a lot of data generated from on-board sensors: speed, power, cadence, and location. Strava - especially the Strava Premium - also has a lot of great tools for analyzing within-ride data, such as time spent in different power zones, or how your times compare on the same section of road in your neighborhood. I wanted to analyze not just the specifics of a ride, but myself as an athlete over time. Basically, I wanted to know if I was getting stronger. I also wanted to gain insights on how to train better.

Here is an example of a Strava activity page.

Screen Shot 2020-05-30 at 10.25.15 AM

How would these numbers trend over time? What could I learn from these trends?

It would be too difficult to go page by page and record these values as I have hundreds of activities. Instead, I scraped the data automatically using the Scrapy python framework. To learn more, please see the Methodology section of this post.

First, the fun stuff! In all of my Strava-recorded history, I have...

Screen Shot 2020-05-30 at 10.21.34 AM

Screen Shot 2020-05-30 at 10.49.13 AM

Now for the more technical analysis. Not all rides are the same. Simply asking if I am "getting faster" over time is not enough. The difficulty of a ride is dependent primarily on gravity, aerodynamics, and distance. To make rides more comparable, I grouped them into mechanical energy spent. Let's take a look at the distribution of energy in kilo-joules.

Distribution of energy in kilo-joules of all rides with bin size = 500 kJ.
kJ_histo

The first four bins capture most of the data. Here is a table to give a rough idea of how kJ translates to mileage and climb.

  rough mileage range rough climbig range
under 250 kJ up to 10 miles 350 ft
250 kJ to 750 kJ 10 to 20 miles 350 ft to 1000 ft
750 kJ to 1,250 kJ 20 miles to 34 miles 1,000 ft to 2,000 ft
greater than 1,250 kJ 34 miles to 62 miles 2,000 ft to 4,000 ft

I can then bin my rides accordingly and see if my average weighted power is trending over time. Power is the best metric because it factors in all the different variables (rate of climb, aerodyanamics, speed, rolling resistance, weight) into a single number.

Average weighted power (watts) over the years by ride total energy (kJ)
watts_year_boxplot

It is important to note that a good training plan is not about always going for maximum wattage. One must train at the different power zones to build different metabolic systems of the body. However, power must on average increase if training is effective.

The A) chart is the noisest. This bin includes the "easy" rides like warm-ups, cool-downs and commutes. I am not surpised since I don't strictly train for power in this energy bin.

B), C), and D) tell a similar story. I have been improving over the years, but my improvement has been tapering off. How I will do in 2020 is yet to be seen.

It is noticeable that I have a lot more data 2017 with a much wider spread. Let's take a look at my frequency of rides.

Riding days per year

days_year

It looks like I have been... slacking. 2017 was a big year for me as I got into racing, and immersed myself in a great cycling community in Oregon.

There is also another interesting observation in 2017: I have a lot of rides in the 250 to 750 kJ range with high wattage: in the 250 to 300+ range. Consider that my functional threshold power at the time was around 294 watts. This is the maximum wattage I could theoretically put out over an hour if I gave my all-out hold-nothing-back effort. Filtering for 2017, 250 to 750kJ with wattage greater than 250 watts, there are 10 points with the following statistics:

  • average distance: 15 miles with sigma = 5 miles.
  • Average moving time: 45 miles.

Directly inspecting on Strava, these were mostly my races on the Portland International Raceway and my FTP test rides.

Since then, I have not been riding as much since my non-cycling life has required a lot of attention. I need to ride more!

A key part of training isn't just about what I do on the bike. It is also about nutrition. How much should I be eating to sustain a cyclist lifestyle?

Calories distribution colored by day of ride

Cal_dist

Most of my rides throughout the week can burn just under 1,000 Calories. But on Satuday, I tend to burn a lot more, averaging at around 1,500 Calories. This will help me plan how much to eat. I hope that I will also make good choices on what to eat.

I can improve my efficiency on the bike by understanding my cadence, which is how fast I spin the pedals. I could go for harder gears to spin slower or go on easier gears to spin faster for the same speed. Understanding my cadence distribution can help me choose what cadence to spin.

Cadence (rpm) distribution

rpm_dist

This indicates I naturally prefer 95 rpm, which is suprising to me because I always felt I was more of an 85 kind of guy. Given this data, I will target 95 rpm on long rides when I am just cruising.

Overall, this is my message for myself:

You are getting stronger, but you need to ride more!

Methodology

If you are a cyclist on Strava and would also like to analyze data, message me. I am happy to help you!

I used Scrapy to web-scrape, and R to analyze and generate graphs. The code can be found on my GitHub repo:

https://github.com/lorenzom21/stravascrape

After logging in to Strava, go to the My Activities section.

https://www.strava.com/athlete/training

Screen Shot 2020-05-30 at 12.13.11 PMWhen you page through your activities, you are not actually going to different web page. This is a single web page with a JSON object feeding into it via a GET request. Every time  you click for the next page, a new object is requested and used to repopulate the data.

I reverse-engineered the GET request by using Chrome's developer tools. I replicated this request in my code, including the headers. Then I scraped all my activities for the activity ID. This is needed because it is used in the URL for the activity page itself. The URL format is:

https://www.strava.com/activities/{activity id}

With activity ID's on hand, I scraped all the activity pages for the numbers I care about. Here is the image again:

Screen Shot 2020-05-30 at 10.25.15 AM

After scraping, I end up with two tables. The first is from scraping the My Activities page, and the second is a collation of the data from the various activity pages. I then join these two tables via the ride id column.

From there, I load up the data in R studio and begin slicing and dicing and charting with R's dplyr and ggplot2 package.

About Author

Lorenzo Mangubat

I enable technologies to scale. My experience ranges from start-up development to high-volume manufacturing. I bring together structured problem-solving, data science, and operational discipline to drive improvements in products, processes and organizations.
View all posts by Lorenzo Mangubat >

Leave a Comment

No comments found.

View Posts by Categories


Our Recent Popular Posts


View Posts by Tags

#python #trainwithnycdsa 2019 airbnb Alex Baransky alumni Alumni Interview Alumni Reviews Alumni Spotlight alumni story Alumnus API Application artist aws beautiful soup Best Bootcamp Best Data Science 2019 Best Data Science Bootcamp Best Data Science Bootcamp 2020 Best Ranked Big Data Book Launch Book-Signing bootcamp Bootcamp Alumni Bootcamp Prep Bundles California Cancer Research capstone Career Career Day citibike clustering Coding Course Demo Course Report D3.js data Data Analyst data science Data Science Academy Data Science Bootcamp Data science jobs Data Science Reviews Data Scientist Data Scientist Jobs data visualization Deep Learning Demo Day Discount dplyr employer networking feature engineering Finance Financial Data Science Flask gbm Get Hired ggplot2 googleVis Hadoop higgs boson Hiring hiring partner events Hiring Partners Industry Experts Instructor Blog Instructor Interview Job Job Placement Jobs Jon Krohn JP Morgan Chase Kaggle Kickstarter lasso regression Lead Data Scienctist Lead Data Scientist leaflet linear regression Logistic Regression machine learning Maps matplotlib Medical Research Meet the team meetup Networking neural network Neural networks New Courses nlp NYC NYC Data Science nyc data science academy NYC Open Data NYCDSA NYCDSA Alumni Online Online Bootcamp Online Training Open Data painter pandas Part-time Portfolio Development prediction Prework Programming PwC python Python Data Analysis python machine learning python scrapy python web scraping python webscraping Python Workshop R R Data Analysis R language R Programming R Shiny r studio R Visualization R Workshop R-bloggers random forest Ranking recommendation recommendation system regression Remote remote data science bootcamp Scrapy scrapy visualization seaborn Selenium sentiment analysis Shiny Shiny Dashboard Spark Special Special Summer Sports statistics streaming Student Interview Student Showcase SVM Switchup Tableau team TensorFlow Testimonial tf-idf Top Data Science Bootcamp twitter visualization web scraping Weekend Course What to expect word cloud word2vec XGBoost yelp