Prasanan Soni | Analytics

Wine Review and Recommendation using the points rated by WineEnthusiast on a scale of 1-100.
Dataset was having raw unstructured data, hence heavy descriptive statistics and exploratory data analysis is done. Word-cloud is used to understand repetitive words and with use of regression model analysis, evaluated elastic net regression (Lasso and Ridge) along with Random forest and Upon getting best accuracy using random forest, recommended best wine as per customer need.

Skin Cancer Diagnostics
Used 300 Images of moles, contains 150 benign and 150 malignant skin images, separated into two different folders. classification techniques to predict device failure for proactive resolution of customer installs.

Network health monitoring using anomaly detection methods (Network Monitoring - Cable Industry)
Used anomaly detection techniques to monitor true health state of network & customer premisis equipment that allows proactive measures to be taken ensuring customer satisfication.

World Wide Car Sales Trend
Used D3.js and tableau public to analyze the world wide car sales and year over year car sales trend for tier1 and tier2 companies. The report clearly shows that steps that GM has taken to target tier 2 countries is good to avoid 2008 slowdown significantly. During Auto market slowdown as well there were markets like Germany, France and Brazil sales was really good. click_here_to_experience_live_version
Developed recommendation & prediction models for transaction reduction within the context of employee digital workplace services (IT Helpdesk)
Collated data from disparate data sources and mined for deeper insights & identified opportunites for transaction reduction. Developed and validated use cases, used text mining for helpdesk service ticket categorization & developed recommendation models for next best action

Regression modeling of aircraft engine health for predictive maintenance.(Aerospace)
Modeled turbine engine performance using operational parameters, demonstrated engine deterioration, measured engine's state & health index to predict maintenance costs.

Modeled aircraft operator use/abuse profile using clustering methods (Aerospace)
Using feature engineered valiables, applied clustering methods to develop models that classify operator behavior using disparate data sources and developed optimization models for for pricing of maintenance service plans and extended warranty support

Developed predictive models for turbine engine subsystems using pattern mining methods (Aerospace)
Developed pattern mining models for engine subsystems that reflect component wear. Developed diagnostics and prognostics algorithms using data driven approach validated by physics based models.

Predicting Borrower Risk using Lending Club’s loan data Augmented with Alternative Data
The current US system of lending is not very good at assessing the risk of borrowers. It relies on a singular value known as a credit score which is biased against people who have had one late payment or people with no credit history. To challenge this approach, publicly available data is used to augument Lending Club's data to predict borrower's risk.

Instacart product recommendations invoked from a web application
The web application uses python as middleware to fetch data from postgreSQL database to drive online ML predictions through a javascript/bootstrap powered front end.

Prasanna Soni