Inspiration
Machine learning is becoming very influential and is being used in many different fields. ML is being used to improve everything from the way we search the web to the way we diagnose diseases. One of the biggest challenges facing ML adoption is the fact that many ML models are black boxes, meaning that it can be difficult to understand how they work and why they make the decisions they do. We wanted to make an effective ML model which goes around this issue.
What it does
Our application uses a machine learning algorithm to predict when certain instruments/assets in a building might need maintenance/replacement. It proactively predicts these times meaning that it prioritizes fixing the machines before they break down. The app also has a lot of customizability and transparency on how it works.
How we built it
We first took the data that was provided and enriched it using it AI. This dataset was used as training set for the ML model. We then also created a test set in a similar fashion and coded an API that can be called to apply the model to the test set. The frontend code takes all this data and makes it intuitive and comprehensible. The user can easily push work orders and customize the priorities on which work orders need to be done first. Budgeting was also a big factor in which the ML model was used for prediction and visualization.
Challenges we ran into
Enriching the data properly was a huge challenge. We used AI to enrich the data, but getting the right prompt was very hard as the enriched data had to be based on reality and be in relation to the features that are already present. Many outliers still had to be removed at the end. Another challenge in developing the ML was getting a proper regression. It took many hours of tuning in order to get a viable model.
Accomplishments that we're proud of
The ML model improved by a huge margin and is actually relatively effective at its predictions. Another thing we are proud of is the intuitive UI. The amount of information and customizability we were able to fit in the app would be very beneficial for the user. It can help the user effectively prioritize work orders and budgeting.
What we learned
We learned a lot about how to handle data and also how to design UI. The UI design especially forced us to push our creativity in order to make it well designed and easy to use while having a lot of features.
What's next for FixMachina
The next step for the app is to create an optional automated process (with easy to place reliable constraints) on automatic work order requests and automatic budgeting.
Log in or sign up for Devpost to join the conversation.