Pratham S.
Data Scientist


1+ years experience

Interested in this profile?




Byte Prophecy

May 2017 - Jun 2018

Data Science Intern

• Developing Data Science models and deploying them at scale using the Big Data Ecosystem
• Developed a Cognitive Science Mathematical model to calculate Variance Impact Analysis
• As a part of the R&D team, developed a Deep Learning model to identify cavities in dental x-ray images


May 2017 - Jun 2018


The Photometric LSST Astronomical Time-Series Classification Challenge Kaggle: PLAsTiCC Astronomical Classification ? Oct2018–Dec2018 Brussels,Belgium
• This project aims to classify simulated astronomical time series data in preparation for observations from the Large Synoptic Survey Telescope (LSST).
• I used cesium library in Python to extract features from the raw time series data and then implemented Gradient Boosting for the classification problem.

Biomedical Image Segmentation
Kaggle: Data Science Bowl 2018 ? Jan2018–Mar2018 Ahmedabad,India
• This project aims to automate nuclei identification in divergent cells images to advance medical discovery.
• I used Deep Learning for this task and implemented the U-Net architecture. U-Net is a Fully Convolutional Network (FCN) that does image segmentation and then predicts each pixel’s class.
I used Keras framework with TensorFlow backend for the implementation.

Dental x-ray Image Segmentation
Byte Prophecy ? Nov2017–Apr2018 Ahmedabad,India
• This project aims to identify the cavities present in a given dental x-ray image.
• The project was done in three phases. During the first phase I only used mathematical morphology. I implemented the sliding windows approach in Python and used OpenCV.
• Realizing the limitations of a simple morphological approach, for the second phase I implemented Random Forests and Gradient Boosting classifiers.The downside to this approach is that the features are hard-coded.
• During the third phase, I implemented my first Deep Learning model.
I used a Fully Convolutional Neural Network which yielded the best results.

Miscelleneous ML Projects
• Quora Question Pairs: Identify question pairs that have the same intent.
• SpamCl assifier:Classify an email as spam or non-spam.
• Movies Recommender: Recommend movies to user based on the ratings provided. Used CollaborativeFiltering.
• Digit Recognizer: Classify handwritten digits using the famous MNIST data.

Newest members that made a profile