PROJECTS

Building Efficiency Comparison, Integrated Building Solutions (IBS)

Objective: Create an algorithm to compare building efficiency for clients of IBS.
Application:
- Worked with real corporate data with over 150,000 + rows of data containing hourly kW and kWh data for 20 buildings.
- Used imputation method to clean data by inserting missing values.
- Created an algorithm to identify building efficiency which aggregated the following three metrics into a uniform criteria for determining building efficiency: Effective Rate (Cost $ / kWh), kWh / Sq. ft, kWh / Population.
- Visualized how buildings rank in efficiency compared to each other and how their building efficiency rank changes over time using a dynamic dashboard in PowerBI. This comparison included overall comparison, weekday vs weekend comparison, and regular hours vs after hours comparison.
Result: Presented the project at IBS to the co-founder and his team and IBS incorporated the algorithm and visuals into their software solution.

Objective: Build a model that outputs the optimal allocation of labels to order for a medical device company who required these labels for their medical devices.
Application: An integer linear programming model was used that divided the amount of labels to order between 10 different suppliers. The model was optimized with respect to minimizing costs while respecting certain constraints like ensuring average quality of labels above 96%, average delivery days below 3 days, and diversification among suppliers.
Result: Successfully ran the integer linear programming model which proposed dividing label ordering between 4 suppliers. This resulted in yearly savings of approximately $26,137 for the company.

Objective: Use machine learning algorithms to predict accurately if a person will earn an annual income of > $50,000 or <= $50,000
Application: A data set with 48,842 records and 15 features (columns) was used and the following machine learning techniques were applied in JMP software: Logistic Regression (normal, forward, backward, mixed step wise), Decision Trees (normal classification tree, bootstrap forest, boosted tree), and Neural Networks.
Result: Successfully ran all models and the best model based on highest prediction accuracy was bootstrap forest (86.29% accuracy). This model was successfully deployed and used to make further predictions on new data.

Objective: Use machine learning algorithms to predict the location (anatomical site) of the primary tumor.
Application: A data set with 339 records and 18 features (columns) was used and the following machine learning techniques were applied in R software: Naive Bayes, Classification Trees.
Result: Successfully ran all models and the best model based on highest prediction accuracy was classification tree (89.71% accuracy). This model was successfully deployed and used to make further predictions on new data.

Objective: Visualize relevant customer data useful for inventory management for different libraries in the city and county of San Francisco.
Application: Built a Tableau dashboard and story that communicated relevant segmented customer data. Visualizations such as treemaps, stacked bar charts, dynamic bubble plots, and pie charts were used to identify patterns in registrations, renewals, and circulations with respect to location of library and age of customer. These trends help the library in managing their inventory accordingly.
Result: Facilitated inventory management for libraries in San Francisco.

Objective: Create a database consisting of select entities for a specific business
Application: Built a relational database for Amazon. Specifically, about ten entities were used relevant to ordering products from Amazon, utilizing MS Access & Oracle SQL Plus
Result: Successfully created the Amazon database, populated it with sample data and queried it to extract relevant information useful for the business.