Looking at data and gaining an insight from them, to make decisions that can affect your organization. This was the core theme of Dr. Ajith P. Madurapperuma’s opening keynote at the “Data Science: An Executive’s Guide”. Organized by the SLASSCOM Technology Forum and held at the Dialog Auditorium on the 7th of February 2018, the guide took the form of a number of keynote speeches and a panel discussion.
What is a data scientist and what is data science?
Kicking things off, Dr. Madurapperuma spoke about who a data scientist is and the knowledge they need. He spoke about the process of data science for decision making. You need to frame the problem. Collect available data, process the data, explore the data and then carry out an in-depth analysis of the data collected. Once all that is done, you need to communicate the results. This involves converting the insights you have gained into an actionable story.
Dr. Madurapperuma then drew examples of a case study of Google where the topic was whether Mangers matter to a team or not. By gathering data and carrying a data analysis, they discovered 8 traits of good managers and 3 traits of a struggling manager. We need people who will understand the data they have, the problem they have and how to solve the problem that they have with the data that they have.
He then spoke about the different types of analytics: Descriptive, Diagnostic, Predictive and Prescriptive. Of these, he said the last two should be given preference. Here you can forecast what might happen and then recommend an action to take based on the forecast.
So what can we do with data science?
Well, you can make recommendations, image/emotion recognition, customized offerings, healthcare analysis, market basket analysis, fraud/crime detection, and prevention etc.
He then spoke about the misconceptions and myths revolving around this field. These include elements such as amounts of investment for DA being high, the removal of human bias, and that data science is like magic and can solve anything. Algorithms are not the key and are not actually failsafe.
Data science from an Executive View
Dr. Srinath Perera was up next. His topic was how to do data science from an executive’s point of view. He spoke about identifying KPIs and then presenting them via a dashboard. When your system KPIs are not functioning as expected, you might see it as an alert or see it on the dashboard. From there, you would want to go into context and see exactly what’s going on.
Dr. Srinath then spoke about Batch processing. This involves storing data, grouping data and then counting data. There are more complicated variations too. For example, there are some values that degrade very fast, such as the stock market, fraud, surveillance, patient monitoring and traffic. While you have details about the data as it happens, it would serve better if you could predict the data and then react within the shortest possible time.
His case study was tracking people with BLE. For that, you would need the data in real time. He then spoke about machine learning and machine learning models. The models are used to build upon and predict the future events. His next case study was about predicting the waiting time at an airport.
For machine learning, there’s classification, regression, clustering. As such ML is used as a prediction tool, an optimization tool, a measuring/approximation tool, an input tool and an automation tool.
He ended his speech by explaining the testing phase of ML models. He derived his examples from WWII fighter planes and how they predicted where to apply reinforced armor by looking at the bullet holes.
Taking things personally
The last speaker for the day was Rashan Peiris, Software Architect at Zone 24×7. His topic was quite interesting. If you do a lot of online shopping, you would have probably come across personalized product recommendations. Well that was exactly what he was talking about. Data science in real life.
Starting off he explained the process of how personalized product recommendations are calculated. He spoke about the various recommendation types such as viewed items, bought items, trending items, etc. and about how they vary, whether by attribute, location, or channel. He went on to the business rules, such as increasing product coverage with processes like backfill, round-robin, and variety. You also have to know whether to implement bury or boost on your products.
But at the end of the day, merely showing recommendations is not enough. You need to evaluate and optimize in order to see the best results. If your algorithm is not performing well to live traffic, that can lead to a drop in profits. He explained that it wasn’t always that easy since there cannot be any expected results out of recommendations. He wrapped up with the big data toolbox and machine learning toolbox, pointing out which technologies are most suitable for what.
It was time for questions
Following this was a Q&A session, where all 3 speakers shed light on problems the audience had. Most people wanted to know how they would integrate into the data science world as people with little to no IT background. The panelists were very helpful, pointing out that basic skills like programming knowledge were very much needed in the current world, but if not statistics and visualization would always be a viable option. The mindset and curiosity are valued more than technical knowledge, they said.
Once the Q&A session finished, the vote of thanks was delivered, with tokens of appreciation being handed out to all the speakers. And with that “Data Science: An Executive’s Guide” came to a close.