Urban Logic
[03b]
Urban Simulation
[Desciption]
A district-level residential mobility model predicting where households in Riyadh relocate over time using three interchangeable approaches — Spatial Interaction Neural Network and Hybrid — and integrating those movements into the broader urban growth simulation.
[Model Objective]
A district-level residential mobility model that predicts where households in Riyadh relocate over time, and integrates those movements into the broader urban growth simulation. Rather than relying on a single method, the model implements and compares three approaches — Spatial Interaction modelling, Neural Network (Deep Learning), and a Hybrid combining both — allowing the user to select the most appropriate method given data availability and prediction goals. Mobility patterns directly reshape the demographic profile of each district, making this sub-model a critical input to any planning or investment decision that depends on knowing where people will live — from public facilities to retail and commercial activity.
PhD Research
2023
Residential Mobility Model
Know more
The model monitors household relocation flows across Riyadh's 150 districts and their downstream effect on district-level demographic profiles. Validated against observed mobility data, the SI model achieved a Pearson correlation of R=0.69 and the Neural Network model reached R=0.76 for predicted destination locations. The hybrid approach combines the spatial structure of SI with the predictive power of NN, offering the strongest overall performance. Sensitivity analysis across mobility methods reveals how much downstream planning decisions shift depending on which movement pattern is assumed.
[Model Components]
Three interchangeable mobility sub-models (SI, NN, Hybrid) · District attractiveness function based on land prices, commercial activity, and building licenses · Household mobility probability engine · Validation framework comparing predicted vs actual destination locations for 2010 and 2015 · Integration layer feeding mobility outcomes into district population profiles
[Model Results]
The model monitors household relocation flows across Riyadh's 150 districts and their downstream effect on district-level demographic profiles. Validated against observed mobility data, the SI model achieved a Pearson correlation of R=0.69 and the Neural Network model reached R=0.76 for predicted destination locations. The hybrid approach combines the spatial structure of SI with the predictive power of NN, offering the strongest overall performance. Sensitivity analysis across mobility methods reveals how much downstream planning decisions shift depending on which movement pattern is assumed.
[Model Components]
Spatial Interaction Modelling (SI) · Neural Network / Deep Learning (NN) · Hybrid Modelling (SI + NN) · Pearson Correlation Validation · Top-k Predictability Testing
[Tags]
["Spatial Interaction", "Neural Network", "Residential Mobility", "ABM", "AnyLogic"]