Loan Prediction

machine learning approaches with various models

Evaluation Results

ROC-AUC

Learning Curves

• SJSU DATA 245 final group project.
• CRISP-DM cycle is applied for project development.
• Apply various ML models to solve this problem: decision tree, logistic regression, ADABoost and gradient boosting.
• Goals: to predict who would qualify for a loan approval based on descriptive features.
• Final model selection: tunned logistic regression model.

• Technologies: Python, Scikit-learn libraries, Google Colab, Supervised Machine Learning Models.
• Category: Data Science

California Real Estate Analytics

Tableau visualization project.

Story 1: Market hotness pre vs post pandemic

Story 2: Market inventory pre vs post pandemic

Story 3: Fast selling market vs slow selling market

Story 4: 2020 vs 2021 market performance

• Look at CA real estate market in 4 perspective:
  • Story 1: How has the pandemic affected market hotness in California?
  • Story 2: How has the pandemic affected market inventory in California?
  • Story 3: Fast selling market vs slow selling market
  • Story 4: California market performance in 2020 vs 2021
• Dataset: Realtor's zipcode hotness and inventory dataset. Link
• Live Demo: Click Here
• Documentation: Click Here

• Technologies: Tableau, Bigquery.
• Category: Data Visualization

Lifeworks NW Projects

Project hightlights

• Strategic KPI Dashboard: represents key company KPIs to executive board. This dashboard tells roll-up stories about how well company is doing using collateral data from multiple sources: finance, CareOregon, Acorn, Netsmart Avatar, medical notes, billing, HR, etc.
• Caseload Management Dashboard: represents all caseload data and visual charts in one place. At drill-down level, dashboard tell all relevant data for a particular episode of a client (all appointments, notes, assessments, treatment plans, insurances, level of care, ASAM level, outreach history, assigned clinician).
• Revenue & Performance Dashboard: represents key factors (roll-up and detailed drill-down) that critical to analyze performances of company, teams and clinicians. Factors include revenue analysis, missed visits/notes analysis, notes/BHAs submission timeliness, group sessions attendance analysis.
• Intake Capacity Dashboard: represent intake capacity (holds vs schedules) of sites, programs and clinicians.
• Medical Services Dashboard: represent 4 key factors (roll-up and detailed drill-down) to medical service team: missed evaluation, missed follow-up visits, evaluation capacity, follow-up capacity.

• Technologies: PHP, SQL, HTML, CSS, Javascript, Netsmart Avatar, Fusion Chart, PowerBI.
• Category: Web Development & Data Analytics

Wildfire Zone Prediction

Predict counties that are likely of wildfire given necessary features.

• SJSU Data 270 individual project.
• Abstract: In California, wildfires have become undeniable environmental threats that lead to catastrophic results on both human lives and the economy. In fact, most fires recorded in California are caused by humans. Therefore, this study aims to introduce a solution to predict future potential fire zones via a web-based interactive map visualization. The goal is to predict potential zones down to county levels that are potential threats to wildfire given necessary factors like human or natural causes, human-related activities, etc. The data used for this study is historical fire data from USDA. We explore Naive Bayes, k-nearest neighbors and one-vs-rest as machine learning approaches to solve this problem. Besides utilizing scikit-learn’s libraries, we also implement Naive Bayes from scratch. The results show that the Naive Bayes model from scratch is the best approach with the highest average accuracy of 0.73 and macro F1 score of 0.35 and therefore will be used to set a starting baseline for accuracy score. In conclusion, with Naive Bayes model we have successfully set up a good accuracy foundation for this problem; with this application, we can not only predict the most potential fire zones with certain factors, we can potentially learn more about fire patterns that are caused by humans. Therefore we can expand this and integrate other alert technologies to notify authorities in order to make better decisions to prevent future incidents.
  
  Keywords: wildfire prediction, predictive visualizations, naive bayes, knn, one-vs-rest, machine learning, multi-class classification.



---

• Dataset: USDA's historical fire dataset. Link
• Applied ML Models:
  • Naive Bayes from scratch.
  • Naive Bayes Scikit learn's GaussianNB.
  • KNN Scikit learn's KNeighborsClassifier.
  • One vs Rest Scikit learn's OneVsRestClassifier (SVC estimator).
• Documentation: Click Here

CA Fire Occurrences

CA Fire Stats

Data Flow Overall

Data Flow Overall By Zone

Naive Bayes From Scratch

Naive Bayes From Scratch (cont.)

Final Results

• Technologies: Tableau, GCP Bigquery, GCP storage, Google Colab, Scikit-learn's ML libraries (GaussianNB, KNeighborsClassifier, OneVsRestClassifier, SVC).
• Category: Data Science

AI Corn Yield & Price Prediction App

SJSU MSDA final project AI web app integrating machine learning and deep learning techniques

Prediction View

Trending and Prediction View

• Technologies: researching, deep learning, machine learning, data mining, data collection, data preparation, AI web development (Heroku, flask, javascript, HTML, CSS), Pytorch, Keras, GCP, Google Earth Engine, Python, project management, technical report writing, Jira
• Category: Data Science

Image Captioning

SJSU Deep Learning Course Project

Merge Architecture

Encoder-Decoder Architecture

Performance

• Technologies: researching, deep learning, Keras, NLP, Sequence Processing
• Category: Data Science

Supply Chain Graduate Technical Intern

Supply Chain | Data Analytics

Corporate Planning Group - Foundry Technology PPA team - Supply Chain Graduate Technical Intern (summer 2022)

• Power-Performance-Area (PPA) Data analytics: automate PowerBI dashboard for Python interactive visualizations of PPA analytic key metrics

• Skills: data analytics, data automation, data visualization, Python

Intel's Projects (2022-2024)