We live in a world of powerful computer hardware, greater availability to an abundance of data than ever before, and ever improving algorithms—all of which contribute to the increased capabilities of machine learning methods and the growing interest in machine learning tools. However, organizations often struggle to chart a successful path from collecting raw data to deploying a workable machine learning model. According to Max Kanter, co-founder and CEO of MIT spinout Feature Labs, part of the problem is that enterprises often approach machine learning projects as research endeavors that end in papers or PowerPoint presentations, which means they fail to positively impact how business operates.
“One of the best ways a company can build their first machine learning model or accelerate their machine learning process, is by focusing on the quickest path to value. Our automation technology helps companies do that by shortening the path from raw data to deployed model while involving key stakeholders in the process,” says Kanter.
Which is exactly what Spanish Bank BBVA achieved using Feature Labs Software. One industry report estimates that more than $100 billion worth of legitimate transactions are incorrectly rejected per year. By applying Feature Labs’ automated feature engineering to over 900 million recorded transactions, BBVA was able to decrease the number of transactions they incorrectly classified as fraudulent by 54 percent. This was accomplished not by creating a new or better machine learning model, but instead by using Feature Labs’ technology to determine better explanatory variables to feed into existing algorithms. “When we looked into the financial impact of eliminating these false positives for BBVA, we found the potential for millions of dollars of savings across all the transactions processed every single day,” says Kanter.
Feature engineering is the process of taking a raw data set and extracting the explanatory variables that are fed into a machine learning model to make the algorithms work. The features, or variables, are used to train the model to make predictions. If we take the BBVA example, the variables extracted might have included client location when using their card, if clients used a chip as opposed to swiping their card, or length of time since the previous transaction. Feature engineering is an essential step in the process to make machine learning algorithms work, but until now it has been time consuming and tedious. Kanter and Feature Labs have developed the most advanced software for automating the process of feature engineering. They call their process Deep Feature Synthesis.
Kanter came up with the idea for Feature Labs while he was a machine learning researcher at MIT CSAIL in the Data to AI (DAI) Group with his future co-founder Kalyan Veeramachaneni, who is a Principal Investigator at MIT’s Laboratory for Information and Decision Systems. They focused on building tools for applied machine learning and data science, particularly as they applied to the real-world problems of their sponsors. What Kanter and Veeramachaneni found was that their biggest challenge wasn’t building accurate machine learning models, but rather the time it took to get to those solutions. “There was a clear shortage of tools in the market to help with the tedious step of transforming and extracting the features to create accurate machine learning models,” he says.
In 2015, Kanter and Veeramachaneni published the results of their research in a paper called “Deep Feature Synthesis: Towards Automating Data Science Endeavors.” They garnered overwhelming industry attention that encouraged Kanter and Veeramachaneni to found Feature Labs. They were later joined by fellow MIT researcher Ben Schreck. One of their first customers was Accenture, the global management consulting and professional services company. Feature Labs software helped them take all of their historical project management information to successfully build a project manager powered by artificial intelligence.
Their successful collaboration with Accenture was a milestone for the young startup, and it gave them the confidence they needed to proceed. Now well-funded, they have continued to prove themselves and the value of their technology, demonstrated in part through mutually beneficial partnerships with Kohls, Monsanto, DARPA, and most recently with Carahsoft Technology Corp.
If a company has questions but doesn’t have the data scientists or resources to answer them, Feature Labs is a great piece of software to help them accelerate that process.
Based on experiences with clients from a wide variety of fields, Feature Labs recently released Machine Learning 2.0. “It’s a new paradigm for developing and creating new machine learning products and services,” says Kanter. It’s a set of seven structured steps that enterprises can follow to translate raw data into a model for rapid deployment and impact.
At the heart of it all is a desire to innovate, to improve the space in which data scientists work, as evidenced by Feature Labs’ open source software, Featuretools. “Creating Featuretools was a labor of love for all of us at Feature Labs. Based on our many years as data scientists in the trenches, we knew that the technology we had built was going to change the way people built predictive models, and we were really excited to make that available to anyone in the world for free.”
Featuretools is proving popular with data scientists everywhere, from people new to machine learning using the tool to build their first predictive models to consultants building models and proof of concepts for their customers.
Feature Labs also provided their software to MIT’s Office of Digital Learning for their big data and advanced analytics course, giving professional learners the opportunity to train their models with the most advanced technology on the market.
Feature Labs has collaborated with a diverse group of organizations across industries, the tie that binds being the desire to adopt machine learning and increase the rate at which they deploy predictive models. “The common thread,” says Kanter, “is interesting raw data sets that have untapped potential. If a company has questions but doesn’t have the data scientists or resources to answer them, Feature Labs is a great piece of software to help them accelerate that process.”