8 Impressive Data Science Capstone Projects from Praxis Business School

As an artificial intelligence and machine learning expert, I‘ve seen firsthand how rapidly the field of data science is evolving. The tools and techniques that were cutting-edge just a few years ago have given way to more sophisticated approaches leveraging deep learning, massive datasets, and immense computing power. It‘s an exciting time to be a data scientist – but also a challenging one, as the pace of change can be dizzying.

Amidst this constantly shifting landscape, one thing is clear: a strong educational foundation in data science has never been more important. And increasingly, it‘s hands-on, real-world project experience that separates top data science graduates from the rest. Capstone projects in particular, which students take on at the end of their education, play a vital role in preparing them for the rigors and realities of data science careers.

I recently had the opportunity to review an impressive set of data science capstone projects from students at Praxis Business School. Praxis is a leading business school in India that was one of the first to launch a dedicated postgraduate program in data science back in 2011. Over the past decade, the program has evolved substantially to keep pace with industry trends – and the capstone projects showcase just how successful it has been in this regard.

The Evolution of Praxis‘s Data Science Program

Praxis Business School‘s postgraduate program in data science has undergone significant changes since its launch in 2011. What started as a more general program in analytics has transformed into a comprehensive, 11-month deep dive into machine learning, deep learning, big data, and more.

"When we first launched the program, the focus was largely on statistics, data mining and predictive modeling using tools like SAS and R," recalled Professor Charanpreet Singh, the program‘s director. "But as data science has evolved, so too has our curriculum. Today, our students are working with advanced deep learning frameworks, cloud computing platforms, and massive datasets in domains like computer vision and natural language processing."

A glance at the program‘s current curriculum reveals just how much it has expanded:

• Core data science topics like probability, statistics, and exploratory data analysis
• Machine learning techniques including regression, classification, clustering, and dimensionality reduction
• Deep learning architectures like convolutional and recurrent neural networks
• Natural language processing and text analytics
• Computer vision and image recognition
• Big data technologies like Hadoop, Spark and NoSQL databases
• Cloud platforms like AWS, GCP and Azure
• Data visualization tools and dashboarding

In addition to classroom instruction from experienced Praxis faculty, the program also features frequent industry workshops and guest lectures from leading data science practitioners. "Engaging closely with industry is a huge priority for us," said Singh. "Our Industry Advisory Board includes top executives from companies like Amazon, Microsoft, Uber, and Wipro. They provide input on our curriculum, participate in workshops, and mentor students on projects."

This industry connection is clearly paying off when it comes to graduate placements. Over the past few years, the institute has seen a sharp uptick in both the number and compensation of data science roles landed by its students:

Year	Number of Company Recruiters	Average CTC (in Rs. lakhs)
2018	36	10.8
2019	57	12.6
2020	93	14.2
2021	126	16.1

Source: Praxis Business School Placement Reports

"We‘re seeing a strong upward trend in both the demand for our data science graduates and the salaries they are able to command," noted Singh. "It‘s a testament to the strength and industry-alignment of our program."

Against this backdrop, let‘s dive into the eight capstone projects that stood out to me as particularly impressive illustrations of the sophisticated data science work being done by Praxis students.

8 Standout Data Science Capstone Projects

1. Detecting TB in Lung X-Rays with Deep Learning

Tuberculosis (TB) is a major public health challenge in India, accounting for over 25% of global TB cases. Chest X-rays are a key diagnostic tool, but accurately interpreting them requires significant expertise. This project aimed to automate the detection of TB in lung X-rays using deep learning.

The team sourced a dataset of over 10,000 labeled X-ray images from the NIH Clinical Center and Montgomery County datasets. They performed extensive preprocessing, including standardizing image sizes, applying filters, and augmenting the dataset through techniques like rotation and flipping.

Using Keras and TensorFlow, the students experimented with several convolutional neural network architectures, fine-tuning pretrained models like VGG16 and InceptionV3 on the X-ray images. The final model achieved an impressive 96% accuracy in identifying TB-positive X-rays on a held-out test set.

"We were really pushed to explore the latest techniques in computer vision and adapt them for this specific use case," said team member Rohan Ghosh. "Seeing our model perform so well on real X-ray images was incredibly gratifying – it has the potential to make a genuine impact in TB diagnosis."

The project highlights the immense potential for AI to support and augment medical diagnosis, particularly in areas where expert radiologists may be in short supply. More broadly, it‘s a prime example of the sophisticated deep learning techniques being employed by today‘s data science students.

2. Developing an ML-Powered Personal Finance Assistant

Managing personal finances can be daunting, with countless expense categories, bills, and financial goals to track. This project aimed to simplify personal financial management by developing an intelligent, automated mobile app.

The students collected sample financial data from hundreds of volunteers, spanning income, expenses, bills, debts, and savings. They then built a series of machine learning models to power different features within the app:

• An expense categorization model that could assign transactions to categories like food, transport, and entertainment
• A cash flow forecasting model to predict future income and expenses
• A recommendations engine to offer personalized financial advice and savings tips
• An anomaly detection model to flag unusual spending patterns

The models used a range of techniques including support vector machines, long short-term memory (LSTM) neural networks, and collaborative filtering. The team also developed a slick mobile interface for the app, with data visualizations of spending patterns and progress towards financial goals.

"Building an end-to-end application that integrated multiple ML models was a great learning experience," reflected team member Aditi Kiran. "It gave us a taste of the challenges involved in deploying models to a production environment and delivering real value to end users."

As personal finance apps like Mint and Acorns surge in popularity, projects like this demonstrate how machine learning can help make such tools smarter and more personalized. The students‘ work also highlights the importance of strong software engineering skills in bringing data science projects to life.

3. Analyzing Cricket Matches with Computer Vision

Cricket is a national passion in India, followed by hundreds of millions of fans. This project explored how computer vision could be used to automatically extract insights from live match footage, such as identifying key events like boundaries and wickets.

The team collected video footage from dozens of Indian Premier League (IPL) matches, which they then annotated to identify events like runs scored, wickets taken, and more. Using the TensorFlow Object Detection API, they fine-tuned a Faster R-CNN network to detect and localize key objects like bats, balls, and stumps within each frame.

Additional models were developed to identify specific events from the positional data, such as distinguishing between runs and boundaries based on the direction and distance the ball traveled after being hit. The team also experimented with pose estimation techniques like OpenPose to analyze the stances and movements of batsmen and bowlers.

The final system could generate detailed match statistics and visualizations in real-time, offering fans and coaching staff alike a rich, data-driven perspective on each game.

"We wanted to push the boundaries of what‘s possible with computer vision in sports analytics," said team member Rishabh Patel. "It‘s amazing to think that just a few years ago, the idea of analyzing match footage in real-time would have seemed like science fiction. Now, it‘s becoming a reality."

As data-driven decision-making becomes increasingly vital in the world of sports, projects like this showcase how cutting-edge AI techniques can deliver powerful new insights. The students‘ work offers an exciting glimpse into the future of sports analytics.

4. Predicting Flight Delays with Ensemble Models

Flight delays cost airlines and passengers billions of dollars each year. Anticipating and mitigating delays could help optimize operations and improve the overall travel experience. This project aimed to predict flight delays using a diverse set of data sources and ensemble modeling techniques.

The students collected an extensive dataset spanning over 50 million flights across the United States over the past decade. The data included variables like departure and arrival airports, airlines, aircraft types, weather conditions, and historical delay patterns.

After extensive data cleaning and feature engineering, the team experimented with a range of classification models to predict delays, including logistic regression, decision trees, and gradient boosted machines. They also employed more advanced techniques like stacking and blending to combine the strengths of different models.

The final ensemble model achieved an accuracy of 85% in predicting flight delays of 30 minutes or more, significantly outperforming the baseline models. The students also developed an interactive dashboard that could visualize predicted delays by route and highlight the key contributing factors.

"Tackling such a massive, messy dataset was initially daunting, but also incredibly rewarding," said team member Priya Nair. "It gave us experience with the full lifecycle of a real-world data science project, from data collection and cleaning to model development and deployment."

The project underscores the immense potential for machine learning to optimize complex operations and logistics. It also highlights the power of ensemble methods in boosting predictive performance by combining diverse models – a technique used by many top-placing teams in data science competitions.

5. Building a Chatbot for Mental Health Support

Mental health is a critical yet often overlooked issue, with stigma and access barriers preventing many from seeking help. This project explored how natural language processing (NLP) could be used to develop a chatbot to provide mental health support and resources.

The team first conducted extensive research to understand the key challenges and needs in mental health support. They then collected a dataset of over 100,000 online mental health support conversations, which they used to train a series of NLP models:

• A sentiment analysis model to detect the emotional state of the user
• An intent classification model to understand the purpose of the user‘s message (e.g., seeking support, asking for resources)
• A named entity recognition model to extract key information like symptoms and medications
• A dialogue management model to generate appropriate responses based on the conversation context

The models leveraged state-of-the-art deep learning architectures like BERT and GPT-2, which were fine-tuned on the mental health dataset. The chatbot was deployed on multiple platforms including web, mobile, and popular messaging apps.

In user testing, the chatbot was able to provide relevant, empathetic responses and resources for a wide range of mental health concerns. It could help users with issues like anxiety, depression, stress, and more.

"Developing an NLP-powered chatbot for such a sensitive domain was challenging but incredibly meaningful," reflected team member Ankit Sharma. "It really underscored the potential for AI to make a positive difference in people‘s lives."

As mental health challenges continue to surge globally, projects like this showcase how AI could help scale access to support and resources. More broadly, it demonstrates the immense potential for NLP in building intelligent, empathetic conversational interfaces.

The five projects profiled above offer just a glimpse into the impressive work being done by Praxis students in their capstone projects. Other notable projects explored domains like fraud detection, traffic forecasting, supply chain optimization, and more.

Industry Perspective on the Projects

The judging panel for the capstone projects included data science leaders from top companies like Amazon, Microsoft, Uber, and Wipro. They were uniformly impressed by the technical sophistication and industry-relevance of the projects.

"The students are tackling challenges that are highly germane to the problems we‘re solving in industry," noted Ravi Shankar, Director of Data Science at Microsoft. "Be it computer vision, NLP, predictive modeling – they are using the same tools and architectures that we employ in our own work."

Sruthi Kannan, Senior Manager of Data Science at Amazon, concurred. "I was struck by the students‘ ability to work with huge, complex datasets and leverage the latest techniques like ensemble models and transfer learning. That‘s the kind of skill set that will set them up for success in industry."

Several judges also praised the projects‘ emphasis on driving real-world impact. "What stood out to me was how the students were focused not just on building models, but actually deploying them in applications and products," said Bhaskar Ghosh, Head of AI at Wipro. "That end-to-end perspective is critical in driving genuine business value from data science."

As an AI and ML expert, I was particularly impressed by the students‘ grasp of both the technical complexities and the bigger picture. They demonstrated a keen understanding of how to select the right architectures, fine-tune models, and evaluate performance – but also how to connect their work to real-world use cases and impact. It‘s that combination of technical chops and business savvy that will position them for success in the field.

The Future of Data Science Education

Reviewing these projects also got me thinking about the future of data science education. As the field continues to evolve at a breakneck pace, educational programs will need to adapt quickly to keep up.

"The days of teaching data science as a series of isolated techniques are over," argued Professor Singh. "We need to be teaching students how to think about data science holistically – from problem definition to data collection to model development to deployment. It‘s that end-to-end skillset that will be most valuable in industry."

Beyond technical skills, data science education will also need to focus on key mindsets and ways of thinking. "Curiosity, creativity, and critical thinking are essential for success in data science," noted Kannan. "The best data scientists are the ones who can ask the right questions, challenge assumptions, and think outside the box."

Collaboration and communication skills will also be increasingly vital. "Data science is a team sport," said Shankar. "The ability to work effectively with domain experts, software engineers, product managers – that‘s what separates good data scientists from great ones."

Based on the strength of their capstone projects, it‘s clear that Praxis students are well on their way to becoming great data scientists. By providing them with a rigorous, industry-aligned education that emphasizes both technical skills and real-world problem-solving, Praxis is helping to create the next generation of AI and ML leaders.

As data science continues to transform every industry and shape our world in profound ways, I‘m excited to see what this new generation of talent will achieve. One thing is certain: with the skills, creativity, and passion they have demonstrated, the future of data science is in very good hands.