Using Data Science to Develop an App for Better Beekeeping: HiveTracks

[MUSIC PLAYING]

JAMES WILKES: My focus for the past dozen yearsor so has been beekeeping technology.And I've gone into that from two different hats.One is as a beekeeper, myself.My family farm-- we run bees.And have a sideline operation producing honey.And then from my professional career, I'm an academic.

JAMES WILKES [continued]: Computer science is my area.And I've been teaching computer science for 26 yearsor so, here at Appalachian State.About 10 or 15 years ago, I beganlooking at how to marry the beekeepingwork that I do with technology.I really had an aha moment in my bee yard,where I was doing bee work.

JAMES WILKES [continued]: Walking up to the beehive and tryingto remember what I had done with that particular hive.And I had a device in my pocket that Iknew that could give me an assist in being a betterbeekeeper.And so I had this vision of we could use technologyto give me information as a beekeeper thathelps me make good decisions.

JAMES WILKES [continued]: Because beekeeping is challenging these days.There's a lot of stressors on bees.And if you don't manage them, then they'remore likely to die.And that's very different than when I was growing upand had bees in my backyard that my dad had.I began thinking about how I could execute on this idea--

JAMES WILKES [continued]: that was the genesis of Hive Tracks.[MUSIC PLAYING]

SULLIVAN WILKES: I use Hive Tracks in our beekeepingoperation to manage useful data--management data to decide what I'm doing on a daily basis.So generally, I'm checking things, probably, once or twicea week at different areas.So I need to know what I did last timeand what I'm going to need to do this time.I also use it for seeing patterns over the years.

SULLIVAN WILKES [continued]: Looking on the past year, our productionand where our colonies were.And how we grew in numbers.Using that to make decisions for next season.

JAMES WILKES: Beekeeping is primarily,an observational activity.So why do you go and look at your bees every so often?Well, it's because you want to evaluate their health.And see if they need anything from you, as the beekeeper.This hive is an example of an hive monitoring devices.And so this hive has a scale under it here.

JAMES WILKES [continued]: And it has two temperature humidity sensors there.I'm going to go to my app and use Bluetoothto pull that data off.And it goes to the phone and then it goes up to the cloud.And this is the BroodMinder technology.There's several in this space that are doing hive monitoring.

JAMES WILKES [continued]: And this is one of the examples.So for hive monitoring, the things that are--one of the easiest to collect and tell us the most--the temperature of the brood nest is one that's very simple.And is a pretty important data point.The weight itself has a lot of applications and information

JAMES WILKES [continued]: that we can glean out of it.Honey flow, winter stores--things like that.The growth of the hive-- we can monitor how quickly they grow.So this hive's sitting on a scale from Solutionbee.And the data that this collects is weight of the hive.And this is a very accurate measure of the weight.

JAMES WILKES [continued]: Within 10 or 20 grams--I think-- is its accuracy.You see, it's got a whole platform herethat the hive sits on.Also, it collects the outside temperature.And what we're trying to capture is events in the hive--biological events that we can't seefrom our observational data.That's really what this monitoring is trying to do.

JAMES WILKES [continued]: Is just enrich our data set and our understanding.And that's in this whole space of the analyticsaround the bees is, we've got observational datathat we can collect.And then there's automatic data that these devices can collect.And they're secondary data sources, as well.Whether the weather or land use that fill in the picture.

JAMES WILKES [continued]: The sensing data that we have in the hive--again, any kind of sensor that you could put in a hiveis a candidate, right.But that is for certain applications,especially for commercial beekeepers.Remote sensing is the gold standardfor what they would want.[MUSIC CONTINUES]

SULLIVAN WILKES: A lot of HiveTracks' datais observational.So when I go into a yard and I perform any kind of management,whether that's harvesting honey, splitting, feeding,I will record that action.Any kind of sensor data, I will then include that as well.And so pairing, like, temperature and humidityand scale data with my observational data,

SULLIVAN WILKES [continued]: I can get a very good idea of what's going on.Generally with, like, honey production--knowing how much weight they're gaining and how many boxes Ihave on the colony--then I can make a decision based on that,if I need to go put more.[INDISTINCT VOICES]

JOSEPH CAZIER: I came into higher tracksabout two years ago as the chief analytics officer.[Joseph Cazier, Chief Analytics Officer, Hivetracks]And my role is to look at the data they've been collectingover the last 8 to 10 years and to try to make sense of thatand to think about how to take our softwareplatform into the future.And that includes what data modifications

JOSEPH CAZIER [continued]: do we need to make to our softwareto get the data we need, and how to leverage that going forward,how to do research on that, and howto go from a data collection system, whichis what it started as, to a decision-guidance system thatwill help HiveTracks help beekeepers make betterdecisions.I think that data standard, the BXML format

JOSEPH CAZIER [continued]: is a critical piece of our strategy going forward.Because then, our data along with everybody else'sthat uses that standard, can be put together to really developdata-driven solutions.

JAMES WILKES: When Joseph joined the team,his job was, how do we look at HiveTracks from a dataanalytics perspective.We really want to go in this directionand achieve these kind of lofty goals of learning about beehealth and affecting really beekeeping around the world

JAMES WILKES [continued]: through the analytics.What we're trying to achieve is to get as good of a datadescription within a 3-mile radius of this beehive as possible.And so some of that data is when the beekeeper goes inand looks, makes observations.And there's certain things that the beekeeper looks for.Is it queenright?

JAMES WILKES [continued]: How many frames of bees are in the hive?What's the brood pattern?Are there any diseases present?How are the food stores?And then the monitoring measures someof the physical characteristics, like your weight,your temperature, humidity, sound.

JOSEPH CAZIER: There's a number of waysthat I use data analytics in HiveTracks.And so the first piece has reallyspent a lot of time spent on data standardizationand trying to get the data in a format that can be analyzed.And that's a difficult, challenging partof the job that's not talked about a lot.But about 80% of the job of an analyst

JOSEPH CAZIER [continued]: is to just get the data ready so you can analyze it.So for example, just today, we werelooking at honey harvest data.And you have some people record it in pounds, some in kilos.That's an easy conversion.But then you have some record it in gallons or in liters.And that's still fairly easy, but youhave to do a volume transformation on that.

JOSEPH CAZIER [continued]: And then if you merge each manufacturer,if they have a sensor, they record it differentlyat the system level.And so you have to translate that to the right version.You have to account for time delays and intervals.You have to account for as much variance as you can.And the more you can control for that,the more you can tease out what's really happening.

PRESTON MACDONALD: Based on the data usingthe demographic statistics, we'remainly appealing to educated usersand understanding that HiveTracks is a data collectionsoftware.I'm a student here at Appalachian State University.My role in the team is to analyze statistics--detailed statistics produced from the big datathat we're pulling from HiveTracks, as well as doing

PRESTON MACDONALD [continued]: data cleaning and figuring out what's useful datato be applied for statistics.[Preston MacDonald, Student, Appalachian State University]To clean the data, I'm using MySQL Serverto pull a data dump from HiveTracks' database.And once I have it pulled up, essentially, I'mgoing through the tables and extracting what informationthat my team members need.And when I start looking at the data,

PRESTON MACDONALD [continued]: I try to figure out what it is that I can quickly filter out.So if there's ranges, for example,for latitude and longitude, filter outwhat's out of that range--also null values.Really trying to make the big cleanups thatare quick to do, especially because we have deadlines.And we can go back later on to do the details.

DATA VISUALIZATION SPECIALIST: Thisis in SAS Viya, which is kind of a cloud-based platform.

PRESTON MACDONALD: Data analytics, weuse SAS to do that.I use specifically SAS Studio.And if I need to visualize anything, I use SAS Viya.SAS Studio provides any SAS coding that you want to do.And essentially, I'm pulling statisticsthat deal with correlations, see howdifferent values affect each other,

PRESTON MACDONALD [continued]: do correlation statistics.I'm also doing significant tests,maybe trying to figure out if there'sa statistical significant difference between things.And all of this will feed into an analysis reportthat I give to HiveTracks.

JOSEPH CAZIER: And so as we clean the data set,we try to predict something, OK.Then you do what we call trustworthy analytics.And that is, you kind of prepare the data in a waythat you divide it up into various setsthat you can analyze.The best data to predict something

JOSEPH CAZIER [continued]: is data from the future.None of us have that because we don'thave data from the future.And so that we can divide it up.And by dividing it up, you can build a modelas a decision tree, a logistic regression, a neural network,or some advanced technique, on a portion of that data set.And then you use another portion of that data

JOSEPH CAZIER [continued]: set to validate your findings.So by splitting your data set into several different ways,you can do that.And we can use it to predict the health of the hive,if there's likely to be trouble.We think we can, in the future, use it to predict crop yields,or at least influence it, and other things, so you

JOSEPH CAZIER [continued]: can optimize those scenarios.And so training the model, that'swhere you split up the data set.And what you do is you'll put the variables in.And the key is you have to have a reliable target variable.And the key pieces is, is that you'retrying to predict this hive did great,this hive died or underperformed in some way.

JOSEPH CAZIER [continued]: And then you look at all the inputs into that.Really, it used to be, in the beginning of computer sciences,that you would hard code and build an algorithm thatwould add up and do statistics.What's different in machine learningis that you're giving the machinethe output and the input and having it build the algorithm.

JOSEPH CAZIER [continued]: And that's kind of how the machine learning works.Rather than the old way, which is you take the inputs,you give it an algorithm, and it gives you the output.Now you're kind of solving that algebraic equation in reverseby having the answer in the inputsand then using that to develop the algorithm that youcan use in the future.[BUZZING]

DATA VISUALIZATION SPECIALIST: My roleis the data visualization specialist.The first visualization that we've started to makeis of location based for our yards within North Carolinafirst, and then moving out to the rest of the world.We did run into a few speed bumps along the way.

DATA VISUALIZATION SPECIALIST [continued]: The data is not always clean.So I've asked Preston to help me out.

JAMES WILKES: So you've got GPS data.This his yard locations, right?

DATA VISUALIZATION SPECIALIST: Yes, sir.

JAMES WILKES: So you had GPS data, and some of that'snot accurate.

PRESTON MACDONALD: Yeah.We had to clean out probably about 8% of the data points,just based on bad lat and long coordinates,as well as just zeroed entries that are outin the middle of the ocean.

JOSEPH CAZIER: We're doing a lot of visualizations.We're doing studies on specific things.For example, earlier we were talking about a softwareanalysis study.I'm also collecting data from third partiesand analyzing it and looking at other secondary sourcesto bring in to our data.But we try to visualize disease vectors with bees.We try to look at usage patterns and health patterns

JOSEPH CAZIER [continued]: over time and over different locations.And what we're trying to do is to really buildwhat we're branding the Genius Hive, whichis a hive that can tell you not just how it's doing,but what it needs to do better.

SULLIVAN WILKES: One of the thingsthat HiveTracks makes possible is the timely receivingof data and decision making.So without HiveTracks, I would normallybe doing a lot of guessing.How many boxes do I need?How strong are these colonies?And what's going on?The hive count fluctuates and goes up and down.And so to know that number makes me know how much feed to buy,

SULLIVAN WILKES [continued]: how much [INAUDIBLE] I'm going to need,and kind of what revenue I can expect for the seasonand for next season, as well.[SOFT MUSIC PLAYING]

JAMES WILKES: This rich, diverse, deep dataset that we could ask questions that have never been possiblebefore--and there's so many variables.It's a dynamic system with weatherand a biological organism.My hope and vision, still, is by having a volume of data, again,you can ask questions that you could not ask previously.

JAMES WILKES [continued]: And you can find threads of commonalityacross a big data set, especiallyin beekeeping, again, with all the variablesthat are going on.

SULLIVAN WILKES: There's about 6 mites in here.So it's about 2 mites per 100 bees.[SCRUBBING]

JAMES WILKES: So that's 2 per 100.

SULLIVAN WILKES: Mm-hmm.

JAMES WILKES: The things that we'reable to do that weren't possible before is reallythe aggregation of data.And by bringing everybody's data together,there are insights that we can gain that were notavailable from individuals.Before, you could bring together small segments of data.But the power in the aggregate data to the community is--

JAMES WILKES [continued]: much more potential there than there is for the individuals.And again, you couldn't even store this much data,previously.[MUSIC CONTINUES][BUZZING]So the importance of tracking bee health,I don't think it can be overemphasized.Sometimes it kind of sounds like a broken record of, oh no,

JAMES WILKES [continued]: the bees are dying, we're all going to die, type message.And sometimes that gets worn out.But honeybees are fundamental to our food systems.And so many of the food supply supply chainsare dependent on honeybees at the very beginningof those supply chains with the pollination of the crop.

JAMES WILKES [continued]: [BUZZING]

JOSEPH CAZIER: Einstein is believedto have said that if the bees die out,mankind will follow two to three years later.And if you think about it, a third of our food--and really, it's 1/3 of the food by volume, 3/4 by type.So 87 out of 113 crops that are commercially grown in the world

JOSEPH CAZIER [continued]: depend on pollination, most of that by honeybees.[MUSIC CONTINUES][MUSIC PLAYING]