Modelling Profits Forecasts for the Russian Banking Sector Using Random Forest and Regression Algorithms

This study is relevant because market uncertainty induces progressively more attempts at making accurate profits forecasts in the banking sector. The scientific novelty of this study lies in the profits forecasts for the Russian banking sector performed using a random forest machine learning (ML) model and a neural network regression model. Regarding technology, the two models are combined into a cognitive model, as they are executed in the same cloud service (Collab) and have a common dataset comprising a training set, scripts and result output. The aim of the study is to build two models: a random forest ML model and a neural network regression model. The dataset used in the random forest ML model and the regression model included data on the performance of the Russian banking sector and some macroeconomic data on the national economy and the stock market for the period 2017–2021. Specifically, the dataset for the models included the following: key rate (%), growth assets (%), overdue loans (%), gross domestic product (GDP, in billions of rubles), RTS index (points), USD rate (vs. RUB), investments in assets to GDP (%), exchange robots (%), capital outflow (in billions of rubles), bank assets (in trillions of rubles), stock accounts (pcs.), and bank profits (in billions of rubles). The practical relevance of this study is evidenced by the fact that the results of the digital profits forecasting for the Russian banking sector can be recommended for real-world use. In building the cognitive model, we used the Python language in the Collab cloud environment. The mean absolute error of the test set for the random forest ML model (DecisionTreeRegressor) was 414.67, which is 61% lower than for the linear regression model (LinearRegression), which had a mean absolute error of 667.65.