Data analysis, Machine Learning

Meghana Ravishankar

https://meghscorpio**.wixsite.com/website https://github.com/meghana1118

www.linkedin.com/in/meghana1811 **********@*****.*** +1-248-***-****

SUMMARY

-Highly motivated Electrical Engineering graduate with passion in Data Science and Machine Learning with 2 years of experience in project work and tutoring.

-Skilled in machine learning topics like Regression, Classification and Clustering.

-Knowledge of Artificial Neural Networks like Back propagation and Convolution Neural Network.

-Familiar with Statistical approaches like Hypothesis testing and confidence intervals, Probability and Bayesian methods.

-Completed Python for Data Scientist track on Data-Camp and certified in Google Analytics for basic proficiency

-Fair knowledge of BI tools like Tableau, Excel and AWS’s EC2 and S3.

EDUCATION

M.S., Electrical Engineering, University of Alabama in Huntsville December 2018

-Coursework: Machine Learning, Introduction to Neural Networks, Big Data Analytics, Digital image processing, Data compression

-UAH Giveback Volunteer

B.E., Electronics & Communication Engineering, Visvesvaraya Technological University May 2016

TECHNICAL SKILLS

Machine Learning: Regression, Classification and Clustering. Neural Networks and Deep learning.

Statistical Methods: Hypothesis testing and confidence intervals, Probability and Bayesian methods.

Programming Language: Python (NumPy, Pandas, Matplotlib, SciKit-Learn, Keras), MATLAB

Database: Basic PostgreSQL proficiency.

Tools: Google Analytics, Tableau, MS Excel, AWS

Soft Kills: Interested in the learning new technology. Active Listener, Quick Learner, Collaborative

GRADUATE PROJECTS

Histopathologic Cancer Detection

EDA of images, Binary Image Classification of histopathologic scans of lymph nodes with 80% accuracy using a Convolution neural network with Tensorflow backend and keras.

EDA, Feature reduction and Classification of Arrhythmia dataset

Features reduction was done using PCA (accuracy of 50-60%) and Random Forest (accuracy of 60-75%). Classification algorithm logistic regression and KNN was used to classify data into 13 classes. [Tool: Python]

Pre-clustering and Clustering of Devanagari Handwritten dataset

K-Means was used to cluster data into predefined cluster number found using pre-clustering Elbow method. The cluster centers of K-Mean were re-constructed for better visualization. [Tool: Python]

Comparison of K-mean and CNN as a feature Extraction method on MNIST dataset

Built a basic Convolution neural network using Keras to reduce features and classify MNIST dataset(18 min, 99% accuracy) to compare with a new model consisting of K-means for feature extraction and KNN(3.8 min for 87% accuracy) for classification to prove K-Means is also a novel and less time-consuming method for feature extraction and Classification. [Tool: Python]

Implementation of basic Machine Learning Algorithms

Libraries used: Python (NumPy, Panda, Matplotlib), MATLAB

-Simple and multiple linear regression, logistic regression models in MATLAB.

-Design and implemented a Kohonen self-organizing map neural net with 2 inputs and 50 cluster units.

-Implemented a Back propagation Artificial Neural Network with one hidden layer to learn logical functions of a XOR gate in MATLAB and Python.

Implementation of inducing False and Pseudo Color in images (MATLAB)

The MATLAB code assigns a false colour to a true colored image (RGB) and pseudo colour to specified intensity of grey in greyscale image.

Bosch Production Line

Predicted internal failures using thousands of measurements and tests made for each component along the assembly line, to enable Bosch to bring quality products at lower costs to the end using numerical data. Feature extracted using Random Forest and use logistic regression for classification with 65% accuracy. [Tool: Python]

Paper Presentations

-3MT presentation for Fast SVD and Random Hadamard matrix

-Atmospheric Correction on Lossy Compression of Multispectral and Hyperspectral Imagery

Undergraduate Project

Low Power Baseband Digital Drop Receiver For WIMAX/4G System

The power consumed by DDC (Digital Drop Receiver) is reduced (from 0.046w to 0.039w) and over all area is reduced (from 62% to 31%) using a CORDIC algorithm. Filter and decimator positions are interchanged to reduce power consumption.

ADDITIONAL EXPERIENCE

Student Specialist III at University of Alabama in Huntsville May 2017-Dec 2018

-Office of Information Technology: Re-imaging of instructor machines and installing a new Operating System on the PC. Triaged and resolved numerous tickets related to Software Issues on the Instructor machines.

-Online Learning: Designed tests and quizzes for online courses on Sociology, Mythology, History and Management. Captioned lecture videos for better accessibility. Designed Posters for events, created online forms for surveys.

Tools used: Canva, RespondUs, Screencast-O-matic, Google forms, Panopto, Canvas, Qualtrics

Contact this candidate