Ishana Shinde

Projects and Publications

• Covid-19 Data Analysis
  • Conducted comprehensive data analysis of an open-source COVID-19 dataset using Azure Data Studio via Docker, aiming to extract meaningful insights and patterns.
  • Developed a series of complex SQL queries to handle data extraction, data cleaning, and transformation processes efficiently, thereby improving the data's quality for further analysis.
  • Generated a range of data visualizations using Tableau, including trend graphs, pie charts, heat maps, and geospatial analysis, effectively communicating the spread and impact of the virus.
  • Designed an interactive dashboard on Tableau to display the generated visualizations, providing a user-friendly tool for users to explore the COVID-19 data at multiple granularities.
  • Ensured scalability and reproducibility of the data analysis process through the use of Docker, making it easier for others to replicate the setup and carry out similar analyses.
  • Tools: SQL, Tableau, Azure Data Studio, Docker

• Demand Forcasting
  • Analyzed time series data using Facebook-Prophet, Decision Tree, Random Forest, and Linear Regression models, aimed at predicting the unit sales of thousands of products across Favorita stores in Ecuador.
  • Performed exploratory data analysis to understand trends, seasonality, and other influential factors affecting sales.
  • Fine-tuned model hyperparameters through iterative testing, boosting model accuracy by 18% and reducing runtime by 30%.
  • Tools: Pyspark, Databricks, Facebook-Prophet, Regression Models.

• American Pets Alive! and Walmart, Return to Owner Hackathon
  • As part of Team Alpha, crafted an innovative solution to aid pet owners in predicting the probable locations of their lost pets, as well as a dashboard to enable shelter workers to visualize animal travel patterns, ultimately optimizing operational efficiencies.
  • Implemented a robust algorithm using Sklearn to predict the possible radius and location of strayed pets based on historical data, enhancing the chances of pet owners reuniting with their lost animals.
  • Utilized Open Street Map and Folium to visualize and map the probable pet locations, providing a user-friendly way for pet owners to navigate and search.
  • Developed a comprehensive user interface using the Flask framework, incorporating HTML, JavaScript, and Flask with Jinja2 plugin to create a responsive and intuitive design.
  • Incorporated a dynamic dashboard in Tableau to display the nearby shelters and potential pet locations, thereby facilitating a smooth navigation experience for pet owners.
  • Tools: Pandas, Pickle, Seaborn, Numpy, Matplotlib, Sklearn, Open Street Map, Folium, Python, Flask Framework (HTML, JavaScript, Flask with Jinja2 plugin), Tableau, Excel

• Prediction of Cognate Reflexes
  • Implemented a State of Art Graph-based Neural Network to predict cognate reflexes for low-level languages, improving language processing capabilities by 40% and reducing error rate by 25% in the Translation team.
  • Developed a novel solution model inspired by Google’s approach; achieved comparable evaluation scores on metrics such as Bleu and Edit Distance, and exceeded industry benchmarks by 20%.
  • Analyzed and optimized data pre-processing workflows, improving data quality by 70%.
  • Tools: Python, Tensorflow, GCNN, CNN, Google Colab, Keras

• Data Mining Projects - CS 584
  • Led diverse Data Mining projects, mastering concepts like Text Classification (Movie Review Classification), Clustering (IRIS and MNIST Image Clustering), Credit Risk Scoring, and Movie Recommendation System.
  • Executed advanced feature extraction and selection techniques, improving the predictive accuracy of models.
  • Conducted in-depth analysis of data mining results, providing actionable insights to stakeholders.
  • Tools: Python, Decision Trees, KNN, K-Means, XG-Boost, Regression Models.

• Component Based Software Development
  • Acquired hands-on experience in full-stack development, harnessing AWS services such as S3, EC2, and IAM to create and host applications.
  • Containerized and deployed applications using Docker and Kubernetes, ensuring resiliency and scalability of the application in Rancher-managed K8 clusters.
  • Configured continuous integration/continuous deployment (CI/CD) pipelines using Jenkins, enhancing the speed and reliability of product updates.
  • Tools: Docker, Kubernetes, Rancher, AWS S3, EC2, RDS, Springboot, Jenkins.

• I'm Something of a Painter Myself
  • Constructed a Generative Adversarial Network (GAN) to translate unique characteristics from one image collection to another without any paired training examples, thereby enhancing the ability to generate synthetic images.
  • Explored various GAN architectures, identifying the optimal setup for the task, and conducted model evaluation using both qualitative and quantitative methods.
  • Tools: Python, TensorFlow, Keras, GAN's.

• Investment Planning and Tax-Automation using Reinforcement Learning
  • Published and authored the paper “Reinforcing Portfolio Management through Ensemble Learning”, detailing the development of a platform for customer assistance in tax and investment planning.
  • Built the platform using Python, Django, and integrated various technologies like Stanford CoreNLP, Custom OCR, and TensorFlow for different functionalities.
  • Leveraged AWS services (S3, Textract, Sagemaker) for cloud-based data storage, text extraction, and machine learning tasks.
  • Tools: Python, Django, Stanford CoreNLP, Custom OCR, TensorFlow, AWS (S3, Textract, Sagemaker).

• Heart Disease Identification
  • Developed a sophisticated health-risk prediction system leveraging machine learning techniques, aimed at estimating the risk of heart disease in patients based on various health parameters.
  • Conducted extensive data preprocessing using Pandas and Numpy, ensuring high-quality data input for model training.
  • Utilized Scikit-Learn to train a robust predictive model, performing hyperparameter tuning to optimize the model for high prediction accuracy and recall.
  • Created an intuitive user interface with Flask for seamless user interaction, enabling patients to conveniently input health parameters and receive risk predictions.
  • Incorporated comprehensive unit tests using the UnitTest framework, ensuring the reliability and correctness of the system.
  • Automated testing of the system's user interface using Selenium, validating system functionality and enhancing the overall user experience.
  • Demonstrated a high level of efficiency in rule discovery, aiding healthcare professionals in early diagnosis and effective management of heart disease.
  • Tools: Flask, Python, Pandas, Numpy, Scikit-Learn, UnitTest, Selenium.

• Password Pattern Recognition Using Keystroke Dynamics
  • Conceptualized and implemented a cutting-edge authentication system leveraging keystroke dynamics to enhance user account security and detect potential intrusions.
  • Devised algorithms to analyze the unique typing rhythm and pattern of a user during password entry, using Python, Pandas, and Numpy, which acts as a biometric identifier for account access.
  • ntegrated OpenCV to add an additional layer of security by capturing an image of potential intruders who attempt to access the account, providing the user with visual confirmation of unauthorized access attempts.
  • Constructed an intuitive user interface using Tkinter, enabling a seamless login experience and providing users with real-time information on account activity.
  • Streamlined the process of handling potential security threats by notifying users of intrusion attempts, thereby ensuring superior account safety.
  • Tools: Python, Tkinter, OpenCV, Pandas, Numpy

• Analyzing US Economic Data - Extracting data from given dataset and analysing the important features to build a suitable dashboard. Displaying the final dashboard on IBM Cloud. This project was a part of the Python for Data Science and AI course offered by IBM.