Dr. Abhinav Sharma
NMIMS University
Yan Ying, Chan
Science & Technology Editor, JMP
Dr. Abhinav Sharma is an Assistant Professor at the School of Business Management, NMIMS University, Mumbai, specializing in operations and data science. With a rich background in quality improvement and management, his work has gained recognition in international journals. Abhinav’s academic journey, which started in mechanical engineering, eventually led him to explore operations management and business analytics. In this interview, he discusses his passion for quality management, the importance of statistics in his career, and how JMP® software has influenced his teaching and research.
Yan Ying Chan: When did you realize that statistics would play a significant role in your career?
Abhinav Sharma: When I first started discussing my Ph.D. work with my professor (Prof. Indrajit Mukherjee), he mentioned that while I had a good background in optimization, I would need to integrate it with statistics for my research. At first, I was surprised and overwhelmed by the idea of integrating statistics with optimization, but I soon realized that it was just a small challenge compared to the bigger problems I needed to tackle. I took some courses and realized that statistics would be an integral part of my career. While optimization used to be my favorite area, I can now say that statistics has become my favorite area, and though I'm not an expert, I have gained the necessary skills for my research.
Yan Ying: Coming from a biomedical science background, I had to take a statistics module during my postgraduate studies, and I really struggled with it. So, as a teacher, what challenges or objections have students had when it comes to learning statistics? And how does JMP help make the subject more accessible for them?
Abhinav: The primary hurdle I see when trying to make students comfortable with statistics is visualization. And by visualization, I don’t just mean plotting graphs. For example, when explaining the central limit theorem, we tell students that the sample average converges to the population average as you collect more samples. But if you don’t show them how this happens visually, it’s hard for them to grasp. The same goes for understanding concepts like confidence intervals or prediction intervals – students have the numbers, but without seeing how it looks visually, they struggle to fully connect the dots. This is where JMP really shines, with its built-in distribution calculators. Whenever I explain a concept, I immediately turn to these calculators. For instance, when discussing the central limit theorem, I use the distribution calculator to show them how JMP draws samples and how the mean becomes closer to the population mean over time. It makes the concept much clearer. The same applies to the law of large numbers.
Previously, students struggled to connect these abstract ideas to real-life scenarios, but JMP offers many real-world examples. Initially, they also found it difficult to use software for analysis, but JMP made that process much easier compared to other software we tried. While there are other options, I found JMP to be the best for my students. The challenge wasn’t that they didn’t know what to do; they just couldn’t fully grasp it. Once I started using visualization and examples from JMP, they began to understand the concepts and how to apply them using the software.
Even with small data sets, they learned how to perform analysis effectively. I also sought out textbooks that integrate JMP examples and outputs, which helped students appreciate statistics more. They began to realize that statistics isn’t just a standalone subject; these techniques are used everywhere. That's when they started enjoying it.
Yan Ying: It's really encouraging to hear that getting the big picture helps students grasp concepts more tangibly. When they can see how things fit together and understand the real-world applications, it makes the material much more meaningful, rather than just working with abstract concepts on paper.
Abhinav: One important point to add is that without providing business context, explanations can fall short. For instance, if I explain a confidence interval by saying the true mean lies between two values, that’s not enough. We discuss real-life cases, like predicting the cost of treatment at a hospital using regression analysis. Every patient will have an estimated cost, but where the true mean lies is shown through the confidence interval. This way, they not only see it visually but also understand it in a business context, which makes the learning experience much more impactful.
To ensure that students apply their knowledge effectively, we include project components in every course. These projects require students to tackle real-life problems. Through these projects, they encounter issues like data quality and realize that data in the real world isn’t as streamlined as what we discuss in class. This approach helps ensure that they grasp concepts both theoretically and practically.
Yan Ying: Could you share an example of two key research questions you’re currently investigating?
Abhinav: Broadly speaking, we focus on developing more accurate, robust, and precise process models using small data sets. This has been a key research area, and I expect it to be something we’ll continue exploring for the next few years, especially with advancements in machine learning.
In many manufacturing settings, you often don’t have access to large data sets. Design of experiments can help with data collection, but industries aren’t going to halt production for us to run experiments. So, most of the time, even if we have historical data, it may be small – sometimes only 50 or 60 data points. The challenge here, which is a broad research question for us, is: how can we develop accurate, robust, and precise prediction models with limited data? This idea ties back to Shewhart’s philosophy – “give me the right data, even if it’s small.” So, we aim to develop models using relatively small but high-quality data, incorporating advanced parametric and non-parametric approaches.
The second key research question is how to account for uncertainties. Prediction models provide an estimated value, but in real-life implementation, there will always be some deviation between the true response value and the estimated one. How can we give decision makers an estimated value while also accounting for possible variations around it? For parametric approaches, we have confidence and prediction intervals to handle this, but for non-parametric approaches, we often rely on sampling techniques, which can be computationally expensive. Our goal is to come up with more efficient methods to handle this.
We also focus on optimizing these models – how can we use prediction models to optimize processes and determine the best process settings? That's another broad area we explore.
Yan Ying: What role does JMP play in your research methods?
Abhinav: I use JMP primarily for regression analysis and hypothesis testing. It’s a key tool for those areas of my work.
JMP has really streamlined our processes. When we were evaluating software options for our university, we had several vendors approach us. Initially, we were a bit biased toward a particular software that I had used during my Ph.D. and master's programs, but after seeing JMP's demonstration, I was really impressed.
The way JMP is designed is incredibly streamlined. It offers advanced statistical outputs that we didn’t find in other software. Plus, the documentation library is outstanding. Everything is clearly explained with proper references, and it constantly incorporates new advancements in statistics. For instance, in regression analysis, JMP includes the latest tests and techniques, which I found fascinating.
One of the biggest advantages of JMP is how intuitive it is. For example, if I want to run a hypothesis test but I’m unsure which one to use, JMP’s Fit Y by X tool automatically recognizes the data type and suggests the appropriate test. It simplifies the process so much, and everything is so fast! In just one window, you can access all the options, and the results are easy to export and share. You can even export your project, import it into another system, and seamlessly continue working. That has been a huge help.
I’ve also used JMP for neural network modeling with my process data, and it was fantastic. Whatever the developers are doing behind the scenes, neural network fitting is incredibly fast – something that would take a minute or two in Python gets done in seconds with JMP. And the best part is that I don’t need to write any code. If I want to check certain metrics, such as plotted root mean square error (RMSE) or error rates, JMP handles it for me automatically.
Another great feature is the K-fold cross-validation – it’s just a simple option to select, and JMP automatically creates the folds. Plus, the model screening platform is a game changer. On my MacBook, it takes about two minutes to run all the models in JMP’s library and show me what works best. It would have taken me a week if I had to code it all in a programming language!
Yan Ying: I totally agree! The fact that there’s no coding required is a huge advantage. Not just in research, but for many companies as well. I’ve heard the same thing – tasks that used to take weeks or even months to figure out can now be done in a matter of seconds with JMP. I noticed you’ve published a few papers, including one on machine learning in manufacturing optimization. Could you explain, in layman’s terms, the business value of machine learning models? What kind of positive outcomes can an organization expect from using this strategy?
Abhinav: Machine learning models provide value in a number of ways, but I’ll focus on two key areas: increased efficiency and improved decision making. In the past, companies would mostly report on what had happened in the past, and that's still common today. There's still a gap when it comes to embracing predictive and prescriptive analytics. This is where machine learning can be a game changer. It helps businesses predict what may happen in the future, such as how different variables, such as your KPIs, are likely to be impacted by various factors.
Machine learning models help by analyzing real data, building models, and then predicting future outcomes. The models essentially capture what's happening and project how things could evolve, leading to better decision making. It saves time, increases efficiency, and ultimately improves decision-making processes.
Let me give you an example. We approached a company where they were experiencing a lot of waste in their manufacturing process. At the start of every shift, workers would come in and try to figure out the best machine settings without really knowing what was optimal. Each new shift would try something different, then check if the output met quality standards, often resulting in wasted time and materials.
Our team studied their process, collected data, and developed a machine learning model tailored to their needs. The machine learning model can be converted to macro, something they could easily use in Excel or another accessible tool. All they have to do is input the process settings, such as temperature and feed rate, and the model would estimate whether the output would meet quality standards. With this approach, workers will have a much clearer idea of the optimal range for process settings. It reduced the trial and error, cutting down on waste significantly. Of course, this can be taken even further with optimization models, but even at this basic level, the results were impressive. They need not to run the process for a while just to find out it isn’t working – they can know in advance which settings would give them the best results.
The real impact is an increase in efficiency, a noticeable reduction in waste, and significant cost savings. Even at the worker level, they will feel empowered because they had more control and confidence in their decisions. They would know, “If I set the process this way, I’m going to meet the quality standards, and my boss is going to be happy because we’re reducing waste.”
That’s a simple but powerful example of how machine learning can improve decision making and operations, even at the frontline.
Yan Ying: So, given the recent advancements in big data and AI, how do you see the field of business analytics evolving in the coming years?
Abhinav: I foresee business analytics becoming increasingly integrated into decision-making processes. The primary role of business analytics is to aid in making better decisions, and this will only become more prominent.
For instance, with tools like ChatGPT available today, you can simply import data, and it will assist with analysis. As AI and analytics continue to advance, they will become an integral part of decision making at all levels – whether you're at a junior, mid, or senior level. More and more people will embrace these technologies, making them essential for effective decision making.
Yan Ying: What challenges should business leaders be prepared for in the coming years?
Abhinav: There are a few key challenges. First, data quality is crucial. Ensuring that the data you have is accurate and representative of the problem you're addressing is essential. It means revisiting basic sampling techniques and making sure that the data accurately reflects the problem you're addressing. A strong grasp of basic statistics is also essential. It's not just about using convenience sampling but making sure your data is a true reflection of the issue at hand.
Another significant challenge, especially in industries like finance, is data governance and privacy. It’s crucial for business leaders to be vigilant about how they handle user data. When developing AI-based solutions that collect user information, you must ensure that this data is secure and not vulnerable to leaks. In the past, we've seen numerous incidents where user emails and passwords were exposed. For instance, AI apps or recommendation systems, like those used by Netflix, gather extensive user data, including viewing habits. Protecting this data from breaches and ensuring it remains confidential are major concerns. Business leaders need to be proactive about safeguarding data and protecting against attacks. While many are aware of these issues, it’s essential to continually focus on data security and privacy to prevent any potential breaches.
A third challenge is the skill gap. As the demand for data literacy grows, it’s clear that we need more people who are proficient in data analysis. Many colleges still fail to provide formal training in statistics, which is crucial for various fields. Douglas Montgomery, in his textbook on design of experiments, highlights the need for engineers to be well-versed in statistics, probability, and experimental design. To address this, curricula for undergraduate, master’s, and even Ph.D. programs should include comprehensive modules on statistics and machine learning. Regardless of the specific domain, having a solid foundation in these areas is essential for the future workforce.
Lastly, while ChatGPT can be both a curse and a boon, it's essential to use it wisely. It has its limitations, and its results are not always perfect. For instance, ChatGPT might provide a quick analysis or summary, but you need to know the underlying theory to evaluate its accuracy. Without a solid understanding of the basics, relying solely on these tools can be problematic.
Yan Ying: Is there anything else you’d like to share?
Abhinav: I want to highlight how impressed I am with JMP’s commitment to providing high-quality software at a reasonable price. Many universities are very conscious of software costs, and JMP offers extensive features at a price point that makes it accessible to academic institutions.
The range of modules JMP provides and the quality of its documentation are exceptional. The references to research papers and formulas used in JMP’s tools offer transparency and confidence in the software’s reliability. This level of detail supports better preparation for classes and helps address student queries effectively.
At our university, NMIMS Mumbai, which has multiple campuses across India, JMP is widely used across nearly all programs and in all the campuses. This widespread adoption speaks to its effectiveness and value. I've even influenced one of the faculty members from another institution to switch to JMP after they learned about it from me. They decided to move away from their previous software and adopt JMP starting this year.
Yan Ying: That’s impressive. You must have a significant influence.
Abhinav: It reminds me of a quote by Steve Jobs: "customers do not form their opinion on quality from marketing." When I used JMP, I realized it’s a high-quality product. This realization, combined with JMP’s affordable pricing, comprehensive modules, teaching aids, and extensive data library, has made it an invaluable tool for me. I no longer need to rely on external data sources like Kaggle, as JMP provides ample data resources. It saves time and enhances the teaching experience, which is why we continue to support and use JMP. That’s the feedback I wanted to share with the JMP team.