- Accuracy not about the model but the data source
- Sharing can deepen the impact of data
BIG data analytics has been touted as the next big thing, but some are still wondering about its effectiveness – an issue that surfaced at a panel discussion on Big Data & Analytics’ Dark Secret, at the second annual What's Next conference organised by Digital News Asia (DNA) in Kuala Lumpur on July 28.
Dr Mubbashir Iftikhar, head of Digital Transformation at KPJ Healthcare, was a proponent, sharing an example of the use of IBM’s Watson cognitive computing technology in cancer treatment in the United States.
By using the technology to parse through data, the probability rate of successfully treating five cancers went up from only 50% to 80-90% six months after use, according to Mubbashir.
“That’s the effectiveness of data – imagine … the cost of using different trials and drugs, all wiped out because of the power of tech,” he declared.
However, the data source is important, according to C.K. Koay, head of customer management in telco Maxis’ Consumer Business unit.
“To do predictive analytics, you basically look at different signals and try to find correlations … accuracy depends not on how well you build the model, but on your data sources,” he argued.
Koay suggested that all the hype around big data analytics (BDA) has made people think that data is easy to get.
But the “biggest challenge is getting clean data – you can’t dump [just about anything] into a machine and get good analytics,” he said, evoking the old IT industry truism of GIGO (garbage in, garbage out).
“You need to know what you are looking for, and then build a model to give you a good score on the things you want to do,” he added.
Getting the right data
Getting to the crux of the matter, Fusionex International founder and managing director Ivan Teh (pic above) warned that BDA is not a magical tool and companies still need to know what question they want to ask.
There also needs to be an adequate amount of data before your analytics model can be accurate. He said, citing the example of a Chinese utility company which wanted to predict and prevent over- or under-supply of power.
“If you were to input three months of data, would you be able to get the right answer? No, because there is spring, summer, autumn and winter … predicting what you need in month No 4 is not going to be accurate,” said Teh.
“If you gave us a full year’s worth of data, is that enough? Perhaps. It is better, but if you gave us three years’ worth of data to insert into the system, it can give you year-on-year comparisons, and take into account other parameters such as population growth,” he added.
Being able to predict downtime for manufacturers is important, and being able to identify symptoms will be key, according to Teh.
“Like a tsunami or a hard disk crashing, there will always be symptoms [that manifest] before these disasters strike, whether it is tremors or vibrations, or whether your hard disk is making weird noises,” he said.
“It is the same with manufacturing – when you want to predict the next downtime, you take a look at the symptoms that happened before, so when you see them appearing again and again, you know that the accuracy in terms of a binary ‘yes or no’ is very high,” he added.
Asking the right question
While the potential for BDA is huge, it still boils down to asking the right question, according to Maxis’ Koay (pic above).
“Even before you start embarking on an analytics project, you have to know what you want to achieve, which would guide you in trying to find the right data or data that could be relevant to you,” he said.
“Building the model is basically math, so when designing it you have training data to feed into the model, then try and apply this to another set [of data], and see if the predictive power is strong enough,” he added.
This use often needs to be iterative – companies need to keep trying or face the consequences of inaccuracy, Koay stressed.
“Most of the time you don’t get it right the first time – you keep on doing it over and over again, changing a variable or two, but once you have a model with a high degree of accuracy, then you can deploy it.
“I don’t think any company would deploy any model that is inaccurate, because it has huge implications.
“Imagine if the model gives false positives, especially for healthcare – that would be scary,” he added.
Fusionex’s Teh said organisations should not stop asking questions during their BDA journey.
“Do you think it is more important to answer the question right or to answer the right question?” he said.
For example, one could answer the question on how to better sell a particular product, but perhaps that would not be the right question if it is not the right product for that market. A better question would be how to sell products that better fit customer requirements, he suggested.
“We say the key thing in BDA is asking the right question – with that problem statement, if we get it right, then we can ensure BDA is not just a tool to answer questions but really works with [decision-makers],” Teh said.
“Because it is a journey, and needs to be brought alive with the right tools, mindset and execution,” he added.
Other What’s Next 2016 stories:
Tycoon Vincent Tan on his costly failures
A digital strategy? You’re behind the times
If it ain’t broke … well, fix it anyway
How to make corporate-startup collaboration work
Focus on SEA’s basic needs, forget Silicon Valley
The generational clash, and sharing vs privacy
Digital disruption not a key concern for Valiram Group
The third digital disruption wave is here
A good VC is polygamous … yeah, you read that right
For more technology news and the latest updates, follow us on Twitter, LinkedIn or Like us on Facebook.