Krannert teams place second and third in machine learning hackathon

Wednesday, December 1, 2021

Teams of Krannert master's students in the Business Analytics and Information Management program placed second and third in November’s Last Mile Hackers Machine Learning Challenge Hackathon sponsored by Tredence, a data science and AI engineering company focused on solving the last mile problem in analytics. The ‘last mile’ is defined as the gap between insight creation and value realization.

As part of the hackathon, teams were asked to prove their mettle with one of three real-life problem statements, including a Retail Recommender Machine Learning problem. Contestants were encouraged to enrich the provided data using publicly available datasets during a two-week long coding round when the students solved the problem. They subsequently shared their solution during the presentation round.

Recommendation systems are used in a variety of industries, from retail to news and media. For example, people who use a streaming service or ecommerce site that has recommendations based on what's been previously watched or purchased have interacted with a recommender system. With the availability of large amounts of data, many businesses are turning to these systems as a critical revenue driver. However, finding the right recommender algorithms can be very time consuming for data scientists.

The second-place Krannert team was made up of Lakshay Vohra, Manideep Sharma, Devansh Batra, and Utkarsh Bajaj. Another team of students from the BAIM program placed third in the same challenge – Pratyusha Gajavalli, Nikhil Katiki, Pavan Ghantasala and Sri Nikhil Bolneyti.

“The problem was to build a solution that utilized AI/ML to predict what a customer will buy next based on their buying patterns,” says Sharma. “Providing business recommendations along with the analytics was an additional challenge. To solve this problem, we needed to first deal with the huge datasets that contained information about the orders, product purchased, and department and aisle details. We decided to solve this systematically.

“Our preliminary step was to understand the data at hand, and then further process it to clean and explore any useful initial trends. Next, we synthesized a master dataset and extracted various key features that could help answer our questions. After engineering the features, we tried several combinations of classification models and selected the best one out of it. In this process, we learned how to hyper-tune the model parameters and implemented it on the go.”

The team members also brought different areas of expertise into the competition, allowing them to combine their analytical results with insights useful for the client. “Through this competition, we truly realized the essence of teamwork and the value of the knowledge that we are receiving at a prestigious educational institution like Purdue,” says Sharma.