A content creator who covers ways to be creative with AI and ML.
2019 is a strange time for transportation. People are riding weird scooters around and using their cellphones to order rides. New York City and London have begun taxing cars that enter city centers in order to cut down on drivers. Transportation alternatives have emerged such as Zipcar, Lime, Bird, and a variety of eBikes or bike shares have increased tremendously proving it’s time to get your rental fleet ready.
While the UK is a long way from California, there are things we can learn from datasets across the world. For one, we can learn from the attributes related to the number of bike rentals how to predict demand. If you’re a Business Analyst, knowing the demand for your product can help avoid shortages or logistical nightmares and cut down maintenance and other costs while increasing usage.
Let’s walk through how to forecast demand using no-code machine learning.
Before using a dataset for predicting demand, you want to identify a specific business problem. In this case of bike share rentals, we want to know what are the type of days a London citizen is most and least likely to rent a bike share. Knowing the problem leads to identifying what kind of data you need to collect.
While there are many different datasets you can collect to identify relationships between usage and variables, the most obvious for cyclists is the weather. Users are more likely to be outside in warmer weather and more likely to use other forms of transportation (Uber, train, bus) to avoid bad weather.
To make your models the most accurate possible, you want to collect the most data you can over the longest amount of time you can. In this case, it’s a good idea to identify patterns in seasonal usage. We used a dataset that collected data over a year-long period to identify what season the bike share rentals are the highest.
Other attributes to get the most accurate ideal day for bike share rentals include:
Some other attributes to collect related to demand could be the geographic location of the rentals, where the user ended the ride, dates of city events such as concerts, games, festivals, etc.
The dataset we used for this use case collected this metadata:
Using this historical data, let’s find some patterns.
Breaking away from the traditional ML process or SQL queries, we take a Natural Language Powered Data Science approach to ask this dataset questions in English. Just upload your dataset, ask questions, get results, repeat.
With the business problem being “What are the type of days a London citizen is most and least likely to rent a bike share so we can provide the right amount of bikes to increase profit and decrease maintenance costs?”
Good questions to ask you data include:
The bike share rental count column is labeled “cnt” in this dataset.
For full transparency, here are the machine learning tech specs used:
From asking the bike share dataset what attributes are proportional to the bike share count numbers, our platform created the ideal day when the count number is the highest:
Feels Like: 29.06°C/85°F
Weather: Partly Cloudy
Wind Speed: 28.46km/h
The days where the number of bike-share rentals was the lowest looks like this:
Feels Like: -4.51°C/23.882°F
Wind Speed: 3.71 km/h
Taking it a step further, we can predict the number of bike shares to be rented on any given day with this data.
For example: At 5 pm on a Summer, non-holiday weekday that is 9.5°C, with a wind speed of 20 km/h, and broken clouds, the predicted count will be 1286.108911 bike share rentals.
Here are the day-by-day prediction results:
Additionally, here is a season-by-season comparison of the number of bike rentals:
0 = Spring, 1 = Summer, 2 = Fall, 3 = Winter
Lastly, here is a comparison in relation to the type of weather:
1 = Clear ; mostly clear but have some values with haze/fog/patches of fog/ fog in vicinity, 2 = scattered clouds / few clouds, 3 = Broken clouds, 4 = Cloudy, 7 = Rain/ light Rain shower/ Light rain, 10 = rain with thunderstorm, 26 = snowfall, 94 = Freezing Fog
Using this process, you now have an unbelievable superpower. You can predict what the bike count will be on any given day as long as you have the historical data.
With Predictive Analytics, you can now plan fleet deployment, promotions, and strategize resources based on weather.
Even though it’s a weird time in the history of transportation, let’s add some level of predictability to it. Before you deploy your fleet, follow this no-code ML process.