Warehouse Construction Progress Prediction for CBRE (https://www.cbre.com/) as part of NYUxCBRE (https://nyudscdatathon2024.com/)
This project aims to predict the stage of progress of warehouses from aerial satellitery images using convolutional neural network (CNN) models and linear regression algorithms. The goal is to achieve between 70% to 60% accuracy in predicting warehouse progress stages and to estimate the number of quarters required for warehouse completion.
Link to slides - https://docs.google.com/presentation/d/1S7SLj6zBwFhO1dV_DiJdLs02rH5djUAqbaNpyj-mwJY/edit?usp=sharing
Link to dataset - https://drive.google.com/drive/folders/1mcrTbJMYEaPSnM5Q1o6_RC8A27pgou_W
Train CNN models to predict warehouse progress stages from satellite imagery. Evaluate model performance using ground truth data. Linear Regression Modeling: Train linear regression algorithms to predict the number of quarters required for warehouse completion. Use input parameters such as progress stages and square footage to predict completion timelines. Integration and Prediction: Combine predictions from CNN models and linear regression algorithms to estimate the total number of warehouses completed within a specific timeframe. Provide insights into the progress of warehouse construction projects.
Data Collection: Gather satellite imagery of warehouses and relevant data such as square footage and progress stages. Model Training: Train CNN models using the collected data to predict progress stages. Train linear regression algorithms to estimate the number of quarters required for completion. Prediction: Use trained models to predict warehouse progress stages and completion timelines. Aggregate predictions to estimate the total number of completed warehouses within a given timeframe.
Bing Maps API code to extract aerial satellite imagery of specific locations using latitude and longitude coordinates.
Contains code for Implementation of Model 1: Simple CNN Model and code for implementation of the Final CNN model
Contains implementation of CNN model with Tensorflow
project notebook containing Linear model solution using Stochastic Gradient Descent with Momentum.
Data: Aerial satellite imagery from Bing Maps; Atlanta supply data related to UC buildings.
Image Specifications: 500x500 pixels; Zoom ratio used: 18.
API: Microsoft Bing Maps (Virtual Earth).
Python 3.x Libraries: TensorFlow, PyTorch, Scikit-learn, Pandas, NumPy Access to satellite imagery data (e.g., Bing Maps API)
Chandana Srinivasa Yatisha ([email protected]) - New York University
Srikar Iyer ([email protected]) - New York University
Unce Shahid ([email protected]) - New York University
Anya Kondamani ([email protected]) - New York University