MAE 600: Machine Learning for Mechanical Engineers Midterm Project Due October 21, 2021 Instructions In this project you will use some of the machine learning methods learned in class to perform complex regression tasks on a real-world dataset. For this project, you will use one of two datasets, described below. The first dataset, provided on Blackboard, is related to building load forecasting. This dataset consists of time series data collected every 15 minutes over a period of three months. The measured quantities include building load (measured by a smart meter) and several weather variables (measured by a local weather station). – To use this dataset, download it from Blackboard. The second dataset, which can be found at https://ashraeobdatabase.com, is related to building occupant behavior. This is a database containing data collected by many researchers while conducting various studies on occupant behavior. – To use this dataset, go to the above URL and select a subset of the available data using the “Export” tab in the navigation menu. Select “Appliance Usage” from the list of behaviors (you may choose others in addition, but you must include Appliance Usage). In the Location and Building lists, select all checkboxes. Finally, select one study from the Study list and export and download the data. This will be your dataset. Please answer the following in a report: Describe the dataset. How many observations does it contain What are the features and target What are the units What is the data resolution Describe any data cleaning and preprocessing you perform. Explore the relationships between each of the input and output variables. Describe your approach to the problem. Will you perform any feature selection What machine learning method(s) will you use Justify your choices. Provide a description of your model architecture and list any hyperparameters. Train and validate your model using your dataset. Tune your model’s hyperparameters and discuss the results after training, validating, and tuning your model(s), using figures and tables where appropriate. Discussion should include topics such as (but not limited to) feature selection, hyperparameter tuning, and overfitting. You should also describe any challenges you faced while developing your model. Please keep the following in mind when preparing your submission: Make sure your report answers all of the above questions. 1 Your report should be no less than 3 pages and no more than 5 pages, double column, single space, Times New Roman, size 12 font, including all plots. If there are many plots, you may include them in the appendix (no page limits for appendix). Your project submission should be in the form of a PDF with supporting PY files. You may include code snippets in your report, but you are required to include your full source code with your submission. If you use data from the ASHRAE OB database for your project, you should also submit your dataset in CSV format. Page 2
