Data Scientist

ABDUL REHMAN

Boston, Massachusetts, ***** 857-***-**** ******.**@************.*** LinkedIn GitHub Available January 2021 to August 2021

EDUCATION

Northeastern University, Khoury College of Computer Sciences, Boston, MA Sep 2019 – Present Candidate for a Master of Science in Data Science, expected graduation: Dec 2021 Teaching Assistant: Computer Science and it Applications Courses: Algorithms, Machine Learning, Data Management and Processing, Causal Inference PES University, Bangalore, Karnataka, India Aug 2015 – May 2019 Bachelor of Technology in Computer Science, Specialized in Data Science TECHNICAL KNOWLEDGE

Languages: R, Python, C, C++, JavaScript, MySQL, HTML, CSS, PHP, Spark Python Libraries: NumPy, Pandas, Sci-kit Learn, TensorFlow, Keras, OpenCV, SciPy, Matplotlib, Plotly, PyTorch, Pyro R Libraries: Ggplot2, Dplyr, Tidyerse, Bnlearn, Boruta, Rose, Caret Technologies:

Techniques:

GitHub, Anaconda (RStudio, Jupyter Notebook, Jupyter Lab), AWS, Microsoft Excel, Tabeau Supervised Machine Learning, Unsupervised Machine Learning, Feature Engineering, Data processing ACADEMIC PROJECTS

Face Detection and Recognition using Deep Learning Sep 2020 – Present

• Developed face detection software in Python, returned bounding boxes of faces in an image with Haar Cascade.

• Leveraged Siamese Network, Convolutional Neural Network, SVM to recognize detected faces using PyTorch and Keras.

• Concluded Siamese Network worked best with 91% accuracy, and did not require retraining on adding employees. Causal Airbnb Analysis. Sep 2020 – Present

• Designed and applied Causal Model in Pyro to analyze online property listings based on location, bedrooms, bathrooms to help potential Airbnb hosts maximize return on investment.

• Created a DAG with 0.98 Global Markov Property to represent factors affecting return on investment from a listing on Zillow.

• Computed the probability of generating profit using the features of the property as input to the DAG. Recommender System Mar 2020 – Apr 2020

• Reduced user-item matrix with PCA, Performed Hierarchical Clustering with cosine distance and complete linkage.

• Carried out Matrix Factorization on clusters, applying SVD to obtain predicted user-item matrix with 0.09 Mean Absolute Error.

• Calculated customer ratings for all products 3 times faster due to hierarchical clustering. Diabetes Prediction Mar 2020 – Apr 2020

• Trained SVM, XGBoost, Random Forest and Logistic Regression models to identify whether a person has diabetes.

• Dealt with highly unbalanced dataset by deploying under-sampling techniques provided by ROSE package in R.

• Selected relevant features using feature importance scores from Boruta package.

• Compared developed models, concluded SVM attains best results with a Sensitivity of 92.6%. Data Augmentation for Images using Style Transfer Jan 2019 – Apr 2019

• Implemented Image Style Transfer model with VGG-16 network in Keras to overcome image data shortage.

• Generated new image by making use of content of one image and style of another, computed Gram matrix to retrieve style and L-BFGS optimization to minimize style and content loss.

• Trained and tested 2 celebrity detection models with 5000 images of 15 celebrities and 10000 augmented images, observed improvement in testing accuracy from 86% to 95% with less overfitting. Sentiment Analysis Oct 2017 – Oct 2017

• Constructed LSTM model in TensorFlow to read comments from amazon reviews and determine sentiments.

• Cleaned data, tokenized sentences to a 5000 words vocabulary and utilized tokens as input to Embedding Layer.

• Compared LSTM model with TextBlob tool, achieved improved AUC score of ~0.9 from ~0.85. WORK EXPERIENCE

Pattern Effects Labs Bangalore, Karnataka, India

Machine Learning Analyst Intern,

Jan 2019 – Apr 2019

• Built LSTM and regression models in Python to look at data from different indicators affecting movement of intraday stocks and judge whether to Buy, sell or hold to maximize profit.

• Improved model performance by carrying out correlation matrix on intraday indicators to select features.

• Attained good models and tested on real time data to calculate profits and evaluate performance.

Contact this candidate