Inspiration
Picking a college is one of the biggest decisions a student can make. Still, most ranking systems only consider academics or institutional prestige. We figured out that a student's happiness on a daily basis mainly depends on their environment, for example, how convenient it is to get a coffee (walkability), the quality of local amenities, how safe they feel at night (safety), or their compatibility with the weather to their lifestyle. With Campus Score, we want to guide the conversation around "Is this a good school?" to "Is this a good school for me?".
What it does
Campus Score ranks universities and colleges not only on academics but also on lifestyle aspects. It takes into account things like walkability, safety, amenities, and weather to come up with a custom score for every school. Since users are also given an option to change settings, the outcome can be very personal.
How we built it
$$
r_{\text{temp}} = \max\left(0, 1 - \frac{\left|T_{\text{avg}} - 18\right|}{25} \right) \cdot 100
$$
where $18^\circ C$ represents the ideal baseline temperature.
Weighted Overall Score
$$
S = \sum_{i=1}^{n} \left( w_i \cdot r_i \right)
$$
where $w_i$ is the user-defined weight for category $i$, and $r_i$ is the normalized sub-score.
Challenges we ran into
Our primary challenge was data normalization. We had to scale disparate units—such as precipitation in $\text{mm}$, temperature in $^\circ C$, and raw counts of amenities—into a unified $0.0$ to $1.0$ range.
To address this, we implemented a Point of Interest (POI) cap:
$$
f(x, c) = \min\left(\frac{x}{c}, 1.0 \right)
$$
where $x$ is the raw count and $c$ is the saturation cap (e.g., $c = 20$ for restaurants or $c = 10$ for hospitals). This prevents extreme values from dominating the score.
Accomplishments that we're proud of
We built a flexible and personalized scoring system that moves away from the old ranking system and puts more emphasis on the user's experience. Through our dynamic weighting system, people can adjust the results according to their liking. Besides, our model is a successful integration of various different data sources from the real world that form one unified evaluation.
What we learned
This project was a real eye-opener regarding the need for user-centric data science. We realized that calling one campus "the best" is a personal thing. Besides, it was pretty hands-on with missing data (NaN), feature normalization, and input scaling via StandardScaler that helped us in training our neural network really well.
What's next for Campus Score
We plan to expand our dataset to include more colleges and additional lifestyle factors such as cost of living and social environment. We also aim to improve our model by incorporating user feedback and refining our weighting system.In the future, we hope to build a full web application that allows students to interactively explore campuses and receive real-time personalized recommendations.
Log in or sign up for Devpost to join the conversation.