Put the date here.

Here's a great place to put a random quote or random post title.

Did someone say multi-dimensional decision surface?

Given all the available information on an area, wading through the valueless information can be a chore. The term segmentation is often applied to similar data in the CRM sphere, where basically you want to see data broken down in segments and compared, e.g., where X > 10 and Y < 10 and Z == 1. Now imagine trying these X, Y and Z parameters are mapped to a multidimensional space, and you have to split this fuzzy sphere of decision points into predictable and unpredictable events, based on the available parameters.

Unfortunately, no-one is willing to let us experiment yet with one of those new-fangled D-Wave (reputed-to-be) quantum computers, so we need to find global minimums and maximums within multi-dimensional data the traditional way… through back-breaking data analysis.

This leads us into a whole area of choosing the right technology for supervised learning, be it support vector machines, naive bayes classifiers, neural networks or some pipeline of multiple methods.

The vagaries of choice.

Borrowing a theme from the good Viktor Frankl, and twisting it completely around, in a world of random chance the only choice we have left is where and when to bet. Exercising our free will, hopefully backed up with some heavyweight statistical analysis, offers at least the illusion of being able to pick a few winners. However, in order to succeed, all forms of external bias should be removed, especially useless crap about a horse having a lucky name, or having a particular appeal towards any given team. As such, anyone wanting to improve their odds should stick to the following rules,

1. The model is right, you are probably wrong.

2. Don’t trust your luck.

3. It’s a marathon, not a sprint.

4. Don’t chase lost causes.

Betting Smart

There is a large, and ever growing market in the area of on-line gambling, with worldwide revenues of upwards of $50 billion USD. However, the market may be considered new, but it still quite traditional, in the sense there are providers (the online bookmakers, casinos, etc.) and clients who place bets with providers. No more different from walking into a bookies shop, scrawling a horse’s name on a betting slip, and giving the cashier some money. And like traditional betting markets, between 95% and 98% of clients end up making a loss in the long term.

The 'uninformed' approach to sports betting.

The ‘uninformed’ approach to sports betting.

Do you see how ‘traditional’ on-line bookmaking can be… like its old-school cousin maybe only 2% of clients ever seeing a profit? It gets worse, as the vast majority of clients see a very minimal return, with maybe 99% of this 2% of clients barely breaking even. Sure, there’s probably a very, very, teeny-tiny number of punters who make significant gains, but lets face it, that won’t be you. The numbers are just too much against you with the way in which the majority of people gamble. Two principal factors serve to limit the vast majority of punter’s revenue possibilities – a lack of analytic capabilities, and a bias towards short term gain.

Taking the latter factor first, colloquially, the long-term strategy is backed up by a story from a friend working in one of the major on-line bookmakers in the UK. One of their on-line clients was consistently making a large profit every quarter, and my friend was asked to analyze their betting behaviour. It turned out this individual was “merely” placing very large sums on almost sure-fire bets. Although the modest return for the large outlay was risky, this individual turned over sufficient revenue to compete with an executive-level salary. In short, this punter was playing the long-game, and it was paying off.

But back to analytics – how do most people place bets in the first place? Sure, some will study the odds, get the occasional tip and try for various savvy doubling-up or doubling-down strategies. However, the dull mass of betting clients will try to guess a lucky horse, have a bias towards a particular trainer, or refuse to bet against their favourite football team. All these inputs only serve to skew a cold analytical evaluation of the odds. While inside information and expert bias may work for well-informed players in the field, jockeys, reserve team coaches, etc., they are unavailable to the average punter in the street.

There is a niche is seems, to provide some guidance to a large swathe of impulsive, happy-go-lucky gamblers who trust instinct over data. I’m “betting” that at least some of them would take a data-driven approach if is was sufficiently accessible, available and workable. So, with that as a taster for the problem area, I’ll leave you to ponder what form such a client-serving analytics capability may take, and what you could do with a fair data-driven assessment of the odds for a match, horse-race or other sporting event. The betting companies build predictive models, so why shouldn’t you?