Yash Shah

Analyzed data anomalies by statistical methods, compared data from other sources, and prepared relevant visualizations for queried 50 states data

Created and scheduled a data pipeline(Cron jobs) to fetch data from FRED warehouse using APIs to GCP BigQuery table

Reviewed Architecture and Generalized the pipeline to collect any data from the FRED warehouse using relevant queries

Developed data processing pipeline to engineer 396 features for producing optimal result on survey responses
Tuned Lasso, Random forest, Gradient boosting prediction models to perform optimally on unseen data

• Created a co-occurrence matrix using memory-based collaborative filtering in python
• Implemented and optimized SVD, KNN for model-based collaborative filtering for movie recommendation

• Analyzed economic KPIs to assess the impact of Covid19 on Canada and developed web-based dashboard
• Web scraped (Selenium) Twitter and Indeed job data to assess the impact on job markets due to pandemic using sentiment analysis and regression analysis respectively
• Benchmarked 2008-09 S&P index data with the recent data to analyze trend similarities for forecasting

• Investigated cost impacts due to scope creep, improper procurement management and stakeholder communication
• Reviewed changes in the project scope, inefficiently delegated tasks and budget estimation impact on schedule

• Implemented simple (SMA) and exponential (EMA) moving average trend indicators for performance analysis
• Applied momentum indicator (MACD) including support and resistance for buy and sell indication in python

• Processed 130k wine review(text) data in Apache spark and Scala for wine variety classification
• Modeled (76% Avg. F1 score) multiclass classifier on imbalanced data using Logistic Regression algorithm

Implemented a gradient boosting algorithm in python for developing machine learning model using LA crime data.
Interactive map integration for data visualization using folium library in python.

The dataset provided (mutiplechoiceResponses.csv) contains the survey results provided by Kaggle. The survey results from 23860 participants are shown in 395 columns, representing survey questions. Not all questions are answered by each participant, and responses contain various data types.
Work involves:

1. Data cleaning with understanding of data, imputing missing values, encoding (label or one-hot encoding)
2. Exploratory Data Analysis
3. Selection of features using… Show more

The dataset provided (mutiplechoiceResponses.csv) contains the survey results provided by Kaggle. The survey results from 23860 participants are shown in 395 columns, representing survey questions. Not all questions are answered by each participant, and responses contain various data types.
Work involves:

1. Data cleaning with understanding of data, imputing missing values, encoding (label or one-hot encoding)
2. Exploratory Data Analysis
3. Selection of features using decisiontree importance and reducing dimensionality with Principal Component Analysis
4. Model implementation using linear regression with lasso and ridge, Random forest and Gradient boosting regressor
5. Model tuning using grid search

In a nutshell, Implemented a machine learning model in python for predicting the compensation based on survey responses of employees in the data science community.

Show less

Analyzed the public opinion about the US airlines on Twitter using python for text and sentiment analysis
Visualized the data using Wordcloud, Seaborn, and matplotlib for the insights.

Steps involved:
1. Data cleaning
2. exploratory analysis
3. feature selection and model implementation for classification
4. models used - SVM, logistic regression, random forest
5. hyperparameter tuning for better accuracy

Improved identification of out of control points using SPC Tools for the Batch production of API and
increased the confidence interval by 0.07% to mitigate the risk.

Analyzed the sensitivity of critical parameters using design of experiments and eliminated cause of
failure in the process.

Improved process by verification and elimination of said cause by design of experiments.

Adapted time-weighted charts for smaller shift detections for GMP with the using of Minitab.

This project is basic implementation of deep neural network - CNN for digit classification

Data set consist of 70,000 digits having 28 x 28 pixels processes :

Dataset import and transformations
Feature engineering
Model creation
Model implementation
Model validation and tuning