I'm getting close to my capstone project at Eastern. I've got a few ideas, but none that seem particularly exciting. I want to build something fun to play with for my capstone, not just useful. I thought that writing here might help stimulate other ideas. Course recommendation system. Take in data from the university about what courses past students have taken and their grades along with their specific degree plan, and based on their individual performance, predict what courses they are likely to be successful in that would also meet the needs of their degree Identify missing persons. Taking the photos and information on unidentified bodies from the National Missing and Unidentified Persons System (NamUs) and comparing them to the database of missing persons in order to find people whose bodies may have been located but not linked to their missing persons case. Predict economic development project success. Using data from state economic development agencies to identify what factors are predictive of project success or failure, and maybe what factors could be adjusted to tip the scale to success (e.g. lowering the amount of jobs to be created or increasing the amount of funding the business owner has to put up) Predict nonprofit failure. Predicting what nonprofits are going to collapse based on data from their Form 990s (extending a PhD thesis by an ODU student: https://digitalcommons.odu.edu/cgi/viewcontent.cgi?article=1045&context=publicservice_etds) Predict volunteer drop out. Analyzing volunteer shift information to determine what factors lead to a volunteer leaving before their 1 year of service. I actually did something like this when I working at a crisis line but it was not very sophisticated Have you seen any cool machine learning projects lately?
Those are pretty cool! I've been kicking around the idea of building a GAN myself, since they can be used for text or image generation. I actually did consider a resume builder that would take in the known variables about someone (e.g. education, work experience, etc.) and the job they are targeting and then generating a realistic-looking resume based on the descriptions of the jobs they did and the descriptions of the job they're targeting.
Do you have some knowledge of CyberSecurity? Machine Learning in CyberSecurity is an emerging topic right now. You can find some open datasets to use.
I don't have experience in cybersec but I appreciate the link, that might be useful for future projects
Lerner's post was full of ideas. Great! Both the article and our man Dustin have referenced "failure predictors." I'm not their customer myself, but I'd be interested in seeing a "failure predictor" for the myriad Government- licensed marijuana stores in Ontario. When these started, they were VERY small in number - I think that number was either 2 or 4 in my city. (I believe that when marijuana retailing licenses were first issued in 2019, there was a total of 25 licenses issued for the entire Province.) Now, they're on just about every street-corner in my town. There were well over 100 in town, halfway through 2021. More, now. I'm sure the pandemic helped them. And of course, the neighbourhood unlicensed dealer - lots of competition still around. Many of the licensed facilities have pretty impressive, professional signage, in what look to be high-rent premises. I think a "failure predictor" for Ontario's licensed marijuana dealers would be interesting. The market has "shakeout" written all over it.
Especially since "grow your own" is also legal, within (what I think are) quite reasonable limits. I've talked to tokers who say it looks like they'll never to have to pay for marijuana again. Another factor for the learning program.
While I'm not a fan of over regulation , some regulation is needed. Once people saturate the market, the margin gets lower and all kind of problems start. US Cannabis might be top shelf, but Canada is the export king and Canadian cannabis growers dominate the export market. The Narco "business" had to adjust , so now instead of Mary the we get Matha (Methamphetamine) smuggled over the border as Mary became more afordible and legal. The side business such as by products Cabanas Oil for medicinal purposes is growing fast, people are investing in it. IT will be interesting to develop such "failure predictor" indeed.
There's no shortage of regulation in the cannabis business, here. The stores are licensed and heavily regulated - the illegal dealers that persist are still illegal, and of course, unregulated. Growing is limited to - I think - four plants per adult - or household, I'm not sure. Not really different to being able to make one's own beer and wine -- which has long been a thing here. With pretty generous limits, I might add. Meth sales aren't a function of pot consumption - or pot legality. Separate markets entirely - one legal, one not. Been here a l-o-n-g time. Scary stuff - untold damage. Meth junkies are the most frightening people I've ever seen - bar none! We've had some homeless 'tweakers' living in the parks (tents) the past couple of years, in spring summer and fall. I don't wanna mess with them! Biggest surprise to me is how incredibly much money the poorest people spend on marijuana. The legal stuff is around $136 Cdn. an ounce. But looking at what they spend on liquor, cigarettes and lotteries - why was I surprised at all? Plenty of things to keep the poor - - poor.
If you can't work (much), due to a medical condition that the plant can help alleviate, it's not surprising that you'd spend $136/oz to get some relief.
People with related medical conditions are not the people I'm talking about. Unless marijuana helps cure chronic laziness, lack of education and often, borderline alcoholism.
Problems like... consumers paying less? Given what I saw in Vancouver, they're not slacking on domestic consumption either. And that's one of the reasons why all of drug prohibition should end, not just cannabis.
As I said, I'm not talking about such people. But - if I did have a condition that could be helped by marijuana - I WOULD NOT pay $136 an ounce, where I live. Or pay anything. The law here says I can grow my own. People (not poor at all) have told me they have had good success with that - and never expect to pay anything for marijuana again. The bulk of the people I'm talking about are recreational WASTERS - pure and simple. Their money is like the beer etc. they consume - in and out .... very quickly.
Is the vote still open on what Dustin should do for the capstone? My vote is the one where you can extend the previous Ph.D. research. I believe you have an interest in pursuing a Ph.D. (and I believe that is a worthwhile endeavor.) Who knows, you may be able to publish something out of your capstone. One thing I noticed about that dissertation (although I did not read it that closely) is that it's focusing a lot on metrics, and not really considering the 'people' aspect in running the non-profits. I don't know, maybe you can extend the research on looking at demographics. Who are these people? What's their prior experience? Did they have sufficient know-how? What about the usual type of demographics like ethnicity or sex? Just thinking out loud. There do appear to be opportunities in extending this research.
So, the proposal I ended up submitting - although not accepted yet - was actually this one: Predict economic development project success. Using data from state economic development agencies to identify what factors are predictive of project success or failure, and maybe what factors could be adjusted to tip the scale to success (e.g. lowering the amount of jobs to be created or increasing the amount of funding the business owner has to put up) However until a couple days ago I was doing a lot of work towards this one: Predict nonprofit failure. Predicting what nonprofits are going to collapse based on data from their Form 990s (extending a PhD thesis by an ODU student: https://digitalcommons.odu.edu/cgi/viewcontent.cgi?article=1045&context=publicservice_etds) Unfortunately there are some major data quality issues with the IRS Form 990 data that I wasn't aware of at the start. I was able to download the 1.8 million Form 990s, clean and extract the 700,000 records for 501(c)(3)'s and put them into Python dataframes for analysis but they frequently had years missing and other data that made it challenging to know when a nonprofit truly failed and stopped updating their Form 990, or if the IRS simply didn't have one on file. This came up so frequently (even in small samples that I used for testing) that I felt I couldn't rely on any conclusions that I came to without a lot more work. I decided that since my goal is to finish the Master's on time, I'll go with the project that has less barriers and then I'll loop around to the IRS data at a later time. I might revisit this IRS project before applying to a PhD though.
