DATA SCIENCE – Urban Delivery Demand GAT Model – 60% complete

[snimle123]

Week 5 pull request – Urban Delivery Demand Prediction project.

This pull request includes the following progress so far:

• Loaded the dataset
• Completed data preprocessing
• Created node features and target nodes
• Created zone index and graph edges
• Prepared feature matrix X and target y
• Converted data into tensors
• Created graph data object
• Split data into training and testing sets
• Defined the GAT model
• Created model object and set optimizer
• Trained and evaluated the model
• Compared actual vs predicted values
• Plotted graphs for results

Current progress: 60% complete

Next steps:

• Improve model performance
• Hyperparameter tuning
• Further evaluation and optimization

[PriyanshuCauleechurn]

Overall:
Nice work on the notebook. The workflow is clear and easy to follow, and I like that the pipeline moves logically from preprocessing the raw delivery data to building the graph structure and training the GAT model. It gives a good foundation for applying graph-based learning to the delivery demand problem.

What was done well:

The notebook is organised in a step-by-step way, which makes the process easy to understand
Good preprocessing choices such as extracting time-based features and calculating delivery duration directly from timestamps
The idea of creating zones as graph nodes is interesting and fits the problem well
The graph construction, tensor conversion, model setup, training, and evaluation are all presented in a clear sequence
The training loss plot and actual-vs-predicted check help make the results more interpretable

Suggestions for improvement:
The node features are scaled before the train/test split, so you may want to check whether fitting the scaler only on the training portion would make the evaluation more rigorous

Conclusion:
Overall, this is a solid notebook with a clear structure and a good modelling direction. Happy to approve. Great work so far Snimle!

[Litxinh123]

Hi Snimle,

Good progress on your submission so far!

Just a few suggestions that could help improve clarity and presentation:

• Include your personal details following the required template (Authored by, Duration, Level, Pre-requisite Skills), you can refer to the attached example
• Add more explanation after each code block to better justify your approach and highlight key insights
• Consider evaluating additional performance metrics (not just loss) to better compare models and identify the most suitable approach

Happy to approve for now. Thanks!