Aritra Basu

In the field of Machine Learning, my business role at Morgan Stanley involves validating Investment management portfolio models and developing benchmark models for those portfolios, which involves machine learning techniques and algorithms such as Random Forest, XGBoost, Natural Language Processing, etc in an agile-based collaborative environment. My day-to-day business activities also include the usage of data science platforms like Databricks, Dataiku, and Snowflake to run scalable analytical languages like PySpark and Scala in Azure/AWS cloud computing platforms.

Few of my major work involves :

1. End to end development of Maximum Entropy Classifier model which serves as a benchmark model for Naive Bayes classifier used as a market standard classifier for NLP based Transaction screening models
2. Independent Benchmarking and Validation of predictive (XGBoost) model to predict attrition of clients in firm-based operations ( Estimated to save 40 % of attrition and 20% at risk attrition, equivalent to $240MM yearly), which involved use of various benchmarking predictive models such as Tree Based models, SVM kernel based models and various Boosting techniques
3. Creating PIP package-based solutions for various modelling techniques used in the team to reduce ~10 business days which was earlier required to develop such predictive models
4. Involved in driving automation projects using development languages such as Flask, Javascript, SQL, etc to track the efficiency metric and capacity analysis of model validation team in Morgan Stanley one - Reduced ∼ 14 business days
5. Involved in Validation of factor-based Asset management portfolio construction models with the likes of BARRA and Aladdin multifactor models which estimates the risk factors associated with securities relative to the market

• 

  Nagasaki University
  June 2022 - October 2022
  During the internship period, my research revolved in the field of detection of brain tumor study. While majority of research in the same study involves detection of brain tumor itself, the innovative part was to bring out the explicit location of the brain tumors in the brain MRI scan.

  In first section of research,mobile net efficient V2 model, a mobile net class TensorFlow model used for object detection, is used for tumor detection from the MRI pictures, while in the second section, efficientNet is used for bespoke object detection along with its image coordinates. Then the images of the tumor are fed into the second part for tumor location detection. For the first section, 1000 photos with a split of 60:20:20 of test:train:validate have been used, with images of brain tumors classified tumors classified as Yes and images of non-tumors classified as No. Following training and validation with the image, the model generates its output on test images with an accuracy rate of 98%. Following tumor detection model, which determines the tumor’s coordinate detection, the data is fed into the efficient net lite 0 model. Label img has been used to annotate photos of tumors with a box around them as test and train datasets for the second model, coupled with an xml file with the tumor’s coordinates. The image with the tumor efficientNet model, is then checked against the outcome is then displayed together with the coordinates and an accurate 96 percent visual depiction of the image of the brain tumor.

• 

  Jaduvpur University
  August 2024 -
  My ongoing research with this institute involves finding suitable background subtraction filters for thermal moving object video captured in the lab. The IRV video, i.e. compatible format of thermal videos, gets processed by Multiple-of-Gaussian filters during the course of research which renders frame-by-frame background subtracted objects. During the course of the study, the concept of frame subtraction was also utilized to visualize and annotate the moving objects getting captured by the Gaussian filter as shown in the right-hand picture. The rectangular box encompasses the moving part of the frame and the ratio of white pixels in the box to the total area of all the boxes in a frame served as a backtesting metric for the research. The metric ranged from 80-95% on average for all the frames of the video, rendering the filter to be a more suitable one for background subtraction of thermal images.

• 

  Detection of location-specific intra-cranial brain tumors
  Shola Usharani, Lakshmanan Rama Parvathy, Gayathri Rajakumaran,A. Basu, Anjana Devi Nandam, Sivakumar Depuru
  Mutations or abnormalities in genes can occasionally cause cells to grow uncontrolled, resulting in a tumor, which is very dangerous. Brain tumors are increasing rapidly, majorly brain tumor cases in the US are projected to rise from 27,000 in 2020 to 31,000 in 2023 at an annual growth rate of 1.5, all the cases are rising because of the detection of the tumors in the late phase. While major research papers on brain tumor detection mainly focus on the detection and classification of the tumors, the given research aims to first detect the tumor using Resnet models on tumor MRI scans. Post successful detection of the tumor, the study plans to determine its precise coordinates and display the tumor and its location in the picture using Efficient Net model.

• 

  Fake News Predictor: A Random Forest-Based Web Application for the Prediction of Fake News on Social Media
  A. Basu, Cinu C. Kiliroor, Ritam Basu & Ritabrata Nag
  The fake news is one of the main concerns nowadays. There are different groups who spread fake news and use that to gain popularity or defame others. In the domain of fake news analysis, our main focus is to help the users to understand whether news is true or fake. The existing models present in the market analyze news based on the text or some time by fact checking in the Internet. These types of model are very time-consuming. Our model is designed based on the identification of the pattern of comment, reaction, and share count on any news. Based on these parameters the proposed model predicts whether the news is real or fake.

In addition to my research pursuits, I hold a deep passion for science outreach. Fostering an appreciation and understanding of science among school children in India is critically important. Engaging with these incredible young minds and kindling their interest in the scientific world bring me immense joy and satisfaction.

I had the honor of working with Youth India Foundation , an NGO based out of Odisha (India), to participate in such a science outreach program, through which I mentored kids from various parts of Odisha State to develop an outreach website that would act as a portal for the ongoing Covid relief campaign in the state. The website helped the organization accumulate sufficient monetary and other forms of relief which they distributed among such rural areas of the state.

I also got the chance to lead a team of full-stack web developers from different parts of the country to develop an online platform known as "E-school", to provide free online reading materials and recorded class videos at a basic education level to anyone free of cost. This website portal enabled thousands of kids from different parts of the country to learn about various national and international content and was also later highlighted by the Times of India on the use of "E-School" platform by the Sundarban education administration on classroom teaching.

I am also a big fan of sports and fitness apart from my academics and research work. I was an integral part of my School Rowing team from 2015-2019 where I acheived a feat of 5 National level Inter School trophies and won various medals for the team. Additionally, I was also a member of my University Handball Team.

You will find my CV here (last update on November 2024).

Commerz III
International Business Park, Oberoi Garden City
Off Western Express Highway, Goregaon (E)
Mumbai 400 063
India

Email: Aritra.Basu1@morganstanley.com