Course Project

For my course project I decided to build a machine learning model to predict the amount of energy a building will use based on characteristics of the building and weather trends. The machine learning model will be trained on a dataset merged from two Kaggle datasets. One dataset includes information about each building including its primary use, total square feet, the year it was built in, and the number of floors in the building. The other Kaggle dataset includes information taken from each site the buildings are located at and includes air temperature, precipitation measurements, and wind speed. This model has two primary use cases. This model is useful as it will allow building owners to compare the expected future energy output given the current characteristics of the building to a theoretical version of the building with modified characteristics. This will allow building owners to make more informed decisions about energy consumption when planning modifications to their buildings.

In terms of methodology, this has been a difficult problem to build a good cross validation strategy for. Typically, in machine learning, datasets can be randomly split into training and test sets. Usually the training set has about 80% of the data and is used to make the model learn patterns in the data, while the test set is used to gauge how well the model is performing. However, for this problem I needed to set up a time series cross-validation strategy. In essence, I had to create the training and test sets in such a way that in any given split of the data, all the training data preceded the test data chronically. This step was necessary because it is not useful to gauge a model’s performance by training on ‘future’ data predicting ‘past’ data. Once, I implemented a working cross validation strategy, I was able to start creating new characteristics about each building from the existing ones and see measurable gains in performance on the test set. I am currently still working on creating more new features, and will soon be moving into feature selection and tuning the hyper-parameters of the model.

Comments
Abdul: https://ae-410-510-ay19-20.blogspot.com/2020/02/b4-ai-in-construction.html
I thought this was a very good introduction to some of the advantages and disadvantages of AI. You are correct in saying AI has successfully decreased human error across a variety of fields. Artificial intelligence can simultaneously keep far more variables in memory than the human mind can and learning can be simulated much quicker than the way in which we learn. This is fantastic for solving complex logic puzzles quickly.

Pritesh: https://ae-410-510-ay19-20.blogspot.com/2020/02/blog-post-4-project-description.html
Gaining more exposure to existing technology sounds like a great fun project to work on. I'm interested to see what type of design you can construct and hope to learn something new from it. I am also new to Revit so understand the amount of learning this will take.

Ina: https://ae-410-510-ay19-20.blogspot.com/2020/01/project-robotics-in-construction.html
I really liked your view on having robots automate the tedious heavy work to give humans more freedom. Metal and machinery is far better at lifting heavy things repeatedly without tiring out than us. By letting the robots take care of this labor intensive side we give humans more time to be creative and innovative. Although, this will initially get rid of some jobs I believe it has the ability to create time for more meaningful and enjoyable activities.