Intelligent Automated System for Managing Application Flows of the Targeted Reward Platform

The aim of this study is to design the architecture of an intelligent automated system for managing application flows of the targeted reward (CPA) platform, integrating four functional modules into a closed-loop system. The objective is to achieve comprehensive automation of management, including fraudulent traffic detection, optimization of click assignment to offers, adaptive bid adjustments, and traffic volume forecasting. The methods used include cascade antifraud based on expert rules and LightGBM gradient boosting; a routing optimizer combining linear programming and ML conversion forecasting; adaptive pricing based on multi-armed bandit algorithms (Thompson Sampling, ε-greedy, UCB1); an ensemble forecaster of traffic volume based on the ARIMA and Holt-Winters models. The scientific novelty of the work lies in the fact that for the first time four modules, heterogeneous in their methodology, have been integrated into a unified sequential pipeline for processing applications of a CPA platform and tested on a common data set. The results obtained on a synthetic data set of 200,000 clicks show that cascade antifraud provides F₁ = 0.787 with Precision = 0.992; LP routing increases the expected revenue by 122%; adaptive bandit algorithms provide an increase of 11...12%; the ensemble forecast provides MAPE = 0.66%. A software prototype is implemented in Python 3.13, comprising 3,298 lines of code, verified by a set of 202 automated tests.