I do machine learning (ML) for a living, but I’ve never applied it to my everyday life. I sold my car when moving to the Bay Area so now I use buses for my commute to work and to get to and from San Francisco. I rarely have better than a 15 minute window of when I think I’ll arrive because bus schedules don’t often hold up to reality. We can use machine learning to change that.

Prerequisites

Data

All ML learning models need data. There’s no getting around it. So it will take some investment on my part before I start seeing any kind of return. The most obvious type of data I can take is time. I’m not going to dedicate my entire hour of commute both ways to collecting data. What I am willing to do is take note of the time when I arrive at stops. The bus makes a loud sound as the doors open that’ll catch my attention.

My morning commute isn’t particularly interesting – I’ve got no stops between where I’m picked up and where I get off:

– Depart: 8:05 AM
– Arrive: 8:57 AM

It will make an easily digestible case – if there’s something wrong with my ML model it’ll show up there first.

My evening commute is more interesting. There are four intermediate stops:

– Depart: 4:17 PM
– A: 4:20 PM
– B: 4:24 PM
– C: 5:21 PM
– D: 5:26 PM
– Arrive: 5:36 PM

From past experience stop C is really closer to 5:11 PM and on average I arrive just a bit before 5:36 PM. I’d like to know when stopping at C within a couple minutes of when I should arrive.

Model

Data isn’t any good unless there’s some way to interpret it. Most ML models require more than a few data points, but there are some statistical methods I can use to analyze the data until I’ve got more. Next week I’ll have four data points to play around with. I’ll see what I can do with them.