Inspiration 📡
Our interest was piqued by the QuadReal challenge, as we wanted to explore air quality sensor data and were interested in the IOT sector.
What it does 🖥️
Our team Ben and Jerry's built a linear regression model, to predict missing values in the IAQ fields of the dataset, as well as detecting anomalies using anomaly detection methods (ANNs) that can adapt to changes in the data fields.
How we built it 🚧
We used Pytorch to train a customized Regression model. We decided on using dropout layers and three hidden layers. For the dataset preprocessing, we decided to use one-hot encoding to treat the features as categorical rather than numerical. We also performed an EDA and tracked results using libraries like pandas, sklearn, numpy, and matplotlib, for visualization and analysis. We used feature engineering to extract relevant features, as well as the regression model to plot and predict any missing values in the dataset before training. We then saved the model for inference, for our validation and test sets (for submission).
Challenges we ran into 🚩
We initially had trouble figuring out the best regression model to use (XGBoost and Random Forest regression were ideas that we ended up scrapping), but we ultimately decided on Artifical Neural Network Regression, as it had the most wiggle room for dynamic-ness as well as being customizable and had a good baseline accuracy.
Accomplishments that we're proud of 💪
We're proud of being able to preprocess the data, build a model, and get a high accuracy result (with a relatively low epoch count) in a short amount of time!
What we learned 🧠
We learned a lot about IoT sensor data, data quality principles, and feature engineering/regression, as we gained a much deeper understanding in ANNs and one hot encoding for conversions between categorical and numerical data. We learned about anomaly detection and using regression for interpolating missing values.
What's next for QuadReal Regression Project CxC ➡️
In the near future, we hope to continue working (hopefully with QuadReal) to improve and expand upon our solution, improving accuracy and model quality.
Log in or sign up for Devpost to join the conversation.