Hype & Prediction of BIG-DATA
Chitro MAJUMDAR
Board Member at RsRL; Founder at AI Ethics & Bias Validation*; Sr. Advisor on Tail Risk Hedging & Risk Quantification et al...
DRAFT-0
As a probabilist I believe that prediction is not really possible, except on nano-intervals and knowing the order book. Together with some knowledge on software used, this could bring something. But in fact this is a clean cheating. Other predictions should only be believed when they are taken as predictions in the past. Data are from the past, we are interested in the future. But data are interesting to find out investor’s behaviour.
I saw a headline a few weeks ago that read: “Big Data is Out; Machine Learning is In.” The post was based on Gartner’s “hype cycle” that showed while big data had big hype in 2014, it has all but fallen off the “hype map” for 2015.
Big banks are now automating their advice for and evaluation of the client’s portfolio. This means machine learning. But we know that there are more pitfalls than results. Machine learning already had good results: car driving assistance. I see "machine learning results" as an assistant in portfolio analysis.
Recently I've seen author Bernard Marr was anxious that, "Does that mean we should start looking for a new job?" ...McDonalds always needs people for hamburger flipping, so don’t worry to find a new job. One thing is true: competition is high, costs must go down and hence banks will downsize their departments and ROI will get more difficult.
DRAFT-1 : Predictive Analytics
How good we are for predicting future? So, you believe Bill Gates saw future? Like many other visionary leaders! Of course their vision (if ever there was one) had less to do with prediction. Bill Gates saw how computer use would develop. He was not the only one. And he had luck or coincidence: IBM chose MS-DOS and so set a standard. The OS was certainly not the best available but when big blue chose it, others followed. And Bill became rich.
Technical Analysis of Predictive Analytics: (linear) regression can be seen in two ways -
1) there is a linear dependence (because of physics or other reasons) but there are errors in the data. Then we use regression to find the best linear approximation (or in more sophisticated applications, non-linear regression)
2) there is some stochastic influence between the data and statistics helps to describe this.
Predictive analysis uses more than regression techniques: they also use cluster analysis, factor analysis (together with regression of course).
{... CONTINUED ...}
Senior Director - Smart products- Gartner | Strategic leadership
9 年Hi Chitro, That's a nice article. One added concern is the application of these machine learning models, without proper understanding and appreciation of underlying assumptions and mathematical implications.
CEO
9 年Waiting with bated breath for next instalment.