- White Wine Quality Prediction Model
A model is created to predict white wine quality. Following an extensive EDA, a wide range of models were implemented and model interpretability was explored. The white wine dataset was retrieved from UCL Machine Learning Repository.
- Adaptive Thermal Comfort Model (Course Project)
Have you ever had an office thermosat fight where you are freezing while your co-worker still complains it's too warm? We have different thermal sensation based on different biologicial and external traits. An adaptive temperature model to predict the optimum room temperature for occupants based on real-time occupant and environment information with logistic regression in Python. Global Thermal Comfort Database II prodvided by ASHRAE is used.
- Day-ahead Hourly Natural Gas Consumption Prediction Model (Course Project)
A model is created to predict day-ahead hourly natural gas consumption for potential natural gas demand response program in Austin, Texas using linear regression and simple neural networks in Python. Pecan Street dataset is used.
- Electricity Time Series Prediction and Clustering Project with Building Genome Dataset 2.
The growth of reliance on the usage of electrical and electronic devices have increased the demand for energy production. Precise energy management from understanding the customer electricity usage patterns and forecast on consumer electricity usage assist utility companies to provide better service, policies and to match supply with demand. This study proposes a clustering-based analysis of electricity consumption using K-means clustering algorithm to categorize consumers electricity usage into different levels based on day types (weekday vs. weekend) as well as typical daily consumption patterns throughout a year. This study also attempts to create an electricity consumption prediction model using K-neighbour regressor algorithm and a simple neural network. K-neighbour regressor algorithm was used to predict 6-month of meter consumption ahead with a year and a half of data points, whereas the neural network model utilizes the past 3 months of meter readings at a given time point to predict the next day electricity meter consumption for 201 days. Building meter reading data from the Building Data Genome 2 (BDG2) dataset is used for this study.