Naveed Ahmed Janvekar has started his career as an engineer at Fidelity investments and has risen to a senior data scientist at Amazon. He has worked with technical teams across industries specialising in early detection of abuse, active learning and social networks. Naveed’s professional journey consists of conducting data science research, speaking at summits and publishing data-science articles on various platforms.
Analytics India Magazine got in touch with Naveed to discuss his data science journey, the fraud prevention space and the must-haves to make it in the industry.
From Fidelity Investments to Amazon data science
Couple of my colleagues at Fidelity Investments were working with data and providing insights for the business. I have developed an inclination towards data when I started to keep a tab on my monthly spending by analysing my transactions. The ability to present raw data into meaningful information led me to pursue data science.
I have decided to get the necessary education to pursue a career in data science and enrolled at The University of Texas at Dallas for a Master’s in Information Science in Data Science. During my Master’s, I interned at Nanigans, an ad-tech startup in Boston. After my graduation, I worked at KPMG, and in 2017 I got an opportunity to work at Amazon in the fraud and abuse prevention team.
Fraud prevention at Amazon
Currently, I work at Amazon as a Senior Data Scientist in the Abuse Prevention team, where I am responsible for implementing machine learning based solutions to detect and prevent any policy-violating entity on the platform. I have been spending a significant amount of time in research and inventions to efficiently use data science/machine learning in abuse detection. Apart from technical work, I spend time working with product managers, software developers, and senior leadership on my organisation’s data science/machine learning vision.
In general, the fraud and abuse prevention space is evolving, which means bad actors constantly try multiple ways to trick or gain access to the platform to commit policy-violating behaviour. Common fraud types that exist across the industry are stolen credit card fraud and account takeovers. In both types, phishing is the common mode attack.
In addition, organisations sometimes must deal with chargeback issues to compensate the victims of credit card fraud. Companies can use data available at the account level at the time of fraud or account takeover to model fraud to predict future risk. Techniques as simple as maintaining a blacklist of stolen credit cards or suspicious IP addresses are also effective in combating such issues in the short term. While organisations can implement sophisticated machine learning models to prevent and detect such attacks – the timing of detection becomes crucial; hence companies can emphasise early detection of such attacks so that the business impact or customer impact is minimised.
Right data science project
The biggest challenge is to get the right kind of projects in the data science field with measurable impact. As data professionals, we can work on many projects, but what makes you really stand out is the value that can be generated or the impact it can have on your customers. I have always prioritised projects that created significant business impact and a positive customer experience in my career. Before starting any data science/machine learning project, I spend significant time scoping out a problem statement and measuring its estimated impact for my company or customers.
Internship
My internship played a big role in my career and gave me the opportunity to work on large datasets and generate insights. Many times, the datasets we work with at school or universities can be clean, but in the real world, datasets can get really messy. During my internship, I experienced challenges in getting the right datasets, munging them, and building a narrative around my findings. As part of my internship, I have published articles such as Facebook Relevance Score: The Magic Number and Facebook Relevance Score: 3 Tactics to Boost Your Ad Effectiveness. The publications were a result of my work on using regression analysis on ad spend optimisation.
Must-have skills
Given the wide scope of a data scientist, a lot of importance is given to being a generalist. But having good scientific understanding of algorithms and the math behind them will give you an edge. Python knowledge, A/B testing skills, metrics development, being able to collaborate with cross-functional stakeholders are super important. Storytelling, building a narrative around your projects, having good product intuition, being able to communicate insights, are also critical.
Constantly researching the latest data science technology, collaborating with peers on various open-source projects and attending conferences are some ways to stay relevant in the field.