Loopwise

• 

  loss function comparison on training data vs validation data

• 

  visualize ride counts after lstm prediction

• 

  hourly ride count for a weekday in a month

• 

  weekday average ride counts for a month

• 

  heatmap of raw_data

• 

  3 different hourly ride count heamaps on the same day

• 

  leaflet map of raw data

• 

  leaflet map of DBSCAN Clusters

• 

  homepage of website

• 

  a loop for a given set of recommended coordinates as per the ML algorithm

• 

  same as above, different coordinates

• 

  option to change radius of how much you would like to loop around, filtering based on that

• 

  same as above

Inspiration

The idea for Loopwise came from seeing the challenges rideshare drivers face while looking for passengers. Drivers often spend significant time idling or driving around aimlessly, leading to lost earnings and wasted fuel. Inspired by traffic heatmaps and optimized delivery routes, we wanted to build a tool that would give drivers actionable insights, showing them where passengers are most likely to need rides. Our goal was to help drivers maximize their efficiency and earnings while minimizing unnecessary driving.

What it does

Loopwise leverages real-time GPS data and historical demand trends to provide rideshare drivers with optimized driving loops, guiding them to high-demand areas at peak times. By following these suggested routes, drivers can increase their chances of picking up passengers without needing to hunt for rides. Loopwise updates routes based on current demand, helping drivers stay in the most profitable areas at all times.

How we built it

We built Loopwise using a combination of Python for data processing, machine learning for demand prediction, and a frontend to visualize route suggestions. First, we processed GPS and historical ride data to identify high-demand locations at various times of day using various python libraries like numpy, pandas, seaborn, folium, and matplotlib (basemap, colors, pyplot, leaflet). We then implemented a recommendation algorithm to suggest loops that would place drivers in these locations efficiently using DBSCAN Clustering and LSTM Time Series models. Finally, we created a user-friendly interface for drivers to view their personalized recommended routes in real time using React.

Challenges we ran into

One of the biggest challenges we faced was managing and processing large datasets of GPS and ride data in real time. Firstly, it was hard to find datasets that has all the features we needed and we ended up using multiple different ones before we zeroed down on the one we liked. Ensuring the route recommendations were relevant to current demand patterns required us to continually update our algorithms. Another challenge was making the route suggestions both efficient and practical for drivers, as we had to balance between accuracy in predictions and ease of following the recommended loops.

Accomplishments that we're proud of

We're proud of creating a system that processes large-scale GPS data efficiently while providing real-time, actionable recommendations for drivers. Our machine learning model successfully identifies high-demand areas and offers accurate routing guidance. Seeing Loopwise help drivers save time and earn more by reducing idle periods has been a huge achievement.

What we learned

Throughout the project, we gained a deeper understanding of GPS data processing, demand prediction, and route optimization. We learned how essential historical data is for forecasting demand, and we also explored algorithms for pathfinding and loop recommendations. Additionally, working with real-world, noisy data taught us the importance of data cleaning and smoothing techniques to deliver accurate, helpful results.

What's next for Loopwise

Next, we want to enhance Loopwise with features like adaptive learning, where the system continuously improves based on driver usage patterns and feedback. We would need to find datasets or data that will constantly update itself so that we can use reinforcement learning to predict demand with a reward function (e.g. how likely it is to get a new passenger based on the location). We’re also exploring integrating weather and event data to make even smarter predictions about demand. In the future, we hope to expand Loopwise beyond rideshare, potentially helping delivery drivers and other mobile professionals optimize their routes and save time.

Built With

• basemap
• dbscan
• folium
• matplotlib
• mplleaflet
• next.js
• numpy
• pandas
• python
• pytorch
• react
• sci-kit
• seaborn
• tensorflow