Development of a Neural Network System for the Real Estate Valuation

In this master's thesis, the problem of estimating the real estate prices using artificial neural networks is considered. Within the framework of this research, the task was to implement software that could solve this problem with a sufficiently high accuracy. To solve this issue, the necessary amount of data was collected from the largest real estate site in Russia. After that, the data was preprocessed for their subsequent use by the neural network. Then, in the programming language Python, using the Tkinter library, software was developed to evaluate 95% of flats in Moscow with a high accuracy. Real and predicted prices were analyzed on the test set. For the optimal set of parameters of the neural network, a training and validation error graph was constructed. A comparison of the determination coefficient with the result obtained in the STATISTICA Neural Network software was made. The received software can be used in the real estate market in Moscow and other cities in Russia.