Hey Jeoff, thanks again for your interesting thoughts. This would probably increase the quality of my analysis, but would triple the amount of mental effort required as well. It’s hard to track these 3 factors every day, opposed to just determining a happiness rating.

But you hit the nail on the head: in Kuwait, I felt like I was working my ass off for something that I didn’t really care about. And my work was influencing a huge part of my life, so that snowball quickly got bigger.

Emailed! 🙂

I’ve been working on an app that I think you would find extremely interested. It is essentially a happiness tracking program that is very much similar to your tracking methods. We should get in touch soon!

Shoot me an email

Hope to hear from you soon!

Hi Lucas,

Thank you so much for your comment! I really appreciate you taking the time to give tips and feedback! 🙂

“In these types of scenarios, often a dataset is split, by random selection, into two segments: one for building the model, one for testing it.”

That makes total sense, yes. This would be a cool approach for the next iteration!

“For instance, find days where only a single factor is listed. Or find days where only positive or only negative factors are listed, and split it between them.”

Again, this should be a very good method for increasing its accuracy!

I really like your recommendations, and am pretty excited to see how they effect the model!

Thanks for stopping by, hope to see you around here more often! 😉

– Be careful about fine-tuning your model too much to track well against past data. You’re testing it against the same data you used to create the model. This can cause your model not to adapt well to new circumstances. In these types of scenarios, often a dataset is split, by random selection, into two segments: one for building the model, one for testing it.

– The damping effect caused by your method of calculating the influence of each factor on your HR could possibly be improved by isolating each effect, if you have enough data for this. For instance, find days where only a single factor is listed. Or find days where only positive or only negative factors are listed, and split it between them. This would also let you test, then, against days with multiple factors of different signs to see if this method really does lead to accurate predictions.

– If you want to get really fancy—and you danced around this point at the end, using only the last 365 days—instead of calculating a single number for the effect of a factor, calculate a regression for the effect of the factor; for a linear regression, it would be y=mx+b, where x would be the date and y would be the factor’s effect in your HR. Or you could do an exponential regression (but don’t over-fit!). Either way, this would allow a factor’s effect to evolve over time.