Geospatial- Artificial Neural Network Technique for Commercial Real Estate Investment Viability Appraisal in Ikeja, Lagos

This study applies geospatial analytics as a component of a broader Geospatial Artificial Intelligence (GeoAI) framework for the appraisal of commercial real estate investment viability in Ikeja, Lagos. Data were collected from 141 registered estate surveying and valuation firms on 143 appraised commercial properties across seven commercial operational areas of the study area. The Artificial Neural Network (ANN) was configured with 41 input neurons, 64 hidden neurons, Rectified Linear Unit (ReLU) activation function, Adam optimizer, which was trained over 50 epochs. Spatial coefficients were derived from Thiessen polygon analysis and were integrated with discounted cash flow variables, as well as property characteristics, as ANN inputs. The ANN achieved a Root Mean Square Error (RMSE) of 238,241.40, Mean Absolute Error (MAE) of 45,104.20, Mean Absolute Percentage Error (MAPE) of 2.1% and R² of 0.3798, outperforming other models like Random Forest, Extreme Gradient Boost, Support Vector Regression, and human appraisers. The ANN-based GeoAI viability mean estimate of ₦676,100.3/m² was closer to the actual mean viability of ₦748,440/m² than human appraisers' mean estimate of ₦843,922.8/m². The results confirm that ANN-based GeoAI models provide a significantly more accurate appraisal than the conventional appraisal method in the complex urban commercial property market.