CBRE Asset Condition Monitor

Inspiration

Our initial plan was to create a web application for Commercial Building Managers that could do two things:

• Show the current status of the assets
• Predict future maintenance needs or potential failures

What we were able to accomplish

• We came up with backend code i.e. the exploratory data analysis, data visualization and prediction model for the asset monitoring app.
• The model can take relevant data like work orders, asset age, repairs etc. and make a prediction of the priority of maintenance.

##How we built it

• Collected the provided fictional data.
• Cleaned irrelevant information and processed the data.
• Implemented feature engineering to add relevant columns.
• Using K-means to perform unsupervised learning and cluster into categories.
• The model gives out the maintenance priority score that is converted into levels i.e. High, moderate, and low.

Challenges we ran into

Finding the right machine-learning algorithm was the most challenging part since there was no target/label data. That is why we had to perform unsupervised learning. We tried other unsupervised learning algorithms as well. However, K-means fitted exactly how we wanted it to be and seemed to give a comparatively better prediction.

Limitations and things that could be improved

We started working on the actual front-end website. However, we couldn't finish it. I will add some pictures to demonstrate how we imagined it to look when we actually implemented the backend code to a user-friendly interface. Limitations:

• Even though it gives a good maintenance priority rating, it does not show potential failure possibility yet. Further work can be done using anomaly algorithms.
• It does not provide notification to the manager since we could work only a bit on the front-end website.
• The model can be tuned to give more accurate with the larger dataset. Evaluation of the model will also help.