An intelligent model for forecasting real estate sales in the primary regional market

Subject of research: an array of data including socio-economic parameters of transactions in the primary housing market of the Chelyabinsk region.

Purpose of research: to create a neural network model for predicting real estate sales volumes with a time horizon of one year based on retrospective data. The predictive model should be easy to set up and operate, as well as provide for an intuitive user interface.

Research methods: preliminary data processing in accordance with the CRISP-DM standard, correlation analysis, normalization, k-fold cross-validation, construction of a fully connected neural network with two hidden layers and a sigmoidal activation function, the L-BFGS optimization method.

Objects of research: forecasting sales volumes in the regional primary housing construction market.

Research findings: a neural network forecasting model has been created using the Loginom Community analytical platform, which has a modular structure, which ensures its versatility. The model was successfully tested on a validation sample, demonstrating a prediction accuracy of 88.33%.