As a part of project in my Business Forecasting class, I created a Long Short-Term Memory model using the tensorflow
package to predict temperatures in Louisiana. These predictions were then integrated with historic temperature data and then used in a model to predict
energy demand. The Jupyter notebook which documents the process of predicting temperatures used can be found
here.
For a project in my Operations Research class, I predicted 300 cars' fuel efficiency (in mpg). The goal of the project was to minimize the Mean Absolute Percentage Error (MAPE).
I was given a spreadsheet that contained a training dataset and a testing dataset. I built a machine learning model using the training set, which was further split into a training/validation set. The MAPE of this model was around 6.1%. I then applied the model to make predictions on the testing set. My final results achieved a MAPE of 6.7%, which was among the best scores ever achieved in the 10-year history of this Operations Research class at LSU.
I first had to process the data that was given to me. The training set did not require to data cleaning, but I recoded some of the columns to make the analysis clearer. I also created dummy variables for the categorical column 'origin'. The testing set required some data cleaning; this was done in the testing_set_cleaner.py file. Many of the functions require "hard-coding" in correct values. The testing set was sufficiently small (at 300 observations) that data-wrangling with a Python script was overkill (but good practice).
In the Jupyter Notebook, I document each step in machine-learning process.
To see my process, see my GitHub page here.
For my Python class in my Master's program, I created a web-scraping script. I decided to scrape Yellow Pages; I scraped restaurants in New Orleans (it is always good to have a list, when you live somewhere with countless places to eat). On my GitHub page you can see the script that I wrote. Click here to see this project.