A team of data scientists associated with Indian Institute of Technology Kanpur predicted Republican candidate Donald Trump’s victory in the US Presidential election six days before the final results.
While every pollster on planet was predicting a clean sweep for Democratic candidate Hillary Clinton, these scientists were so confident of their model that they sent a formal invitation to President Donald Trump on November 3, to be chief guest at IIT Kanpur’s Annual Technical Festival in capacity of President of United States.
Sanjeev Newar, founder of 4Front Analytics incubated at IIT Kanpur, said, "We used an ensemble of statistical methods, machine learning techniques and simulations to create an envelope of victory scenarios for each candidate. Trump trumped most of the litmus tests and hence we concluded that he will win unless 230 million US voters decide to change their minds suddenly.
"The methods used by almost every pollster have some fundamental flaws. First, they ignore the fact that voting is essentially a secret right. Conducting surveys and projecting sample results to entire population just magnifies the fundamental flaws of your study design and data sources," explained Newar whose risk management solution Aarsh is listed in Top 20 of world by CIO Review, USA.
"Further they do not conduct sensitivity analyses and completely ignore model risk.
Also Read
"To get right data, we performed sentiment analysis on social media data that we meticulously collected from various pro and anti groups of both camps. We focussed more on the comments and reactions to extract hidden patterns that voters would not reveal in poll samplings, especially if the candidate they support evokes strong sentiments.
"In fact we found that the social media censure of Trump only helped him. He was gaining strengths not after the debates but after most surveys announced him as loser in debates. The allegations regarding misogyny increased his popularity among women, if data is to be believed. He came across as a straightforward and genuinely apologetic person who speaks the general American language.
"Our models predicted a range of 280 to 310 for Trump. Anything less than that did not make sense. So we decided to invite him officially for the largest Technical festival of Asia happening in March 2017," he added
"It was a tremendous risk since we were going against every prediction on US elections made by experts who had spent millions. But we were so convinced with the model framework that it appeared a certainty. So we just went ahead, sent an invitation and tweeted him," said Dr Mainak Das, faculty coordinator for Techkriti 2017, IIT Kanpur.
4Front provides cloud-based big data and analytics solutions to businesses. It has recently launched a public beta of their cloud offering called Galton.
"Prediction is both an art and science. You must be able to follow systematic process to avoid personal biases, build cross-validations, and extract hidden patterns from proxy data," said Dr Deepu Philip, smarts systems faculty at IIT Kanpur and also a mentor to 4Front Analytics.