The bottom cycle in supercritical carbon dioxide (sCO2) recompression Brayton combined cycle (RCBC) effectively recovers substantial waste heat for electricity generation, enhancing energy utilization. The recovery efficiency is significantly influenced by the bottom cycle configurations and specific working conditions. Besides, implementing these bottom cycles introduces challenges related to increased system dimensions and design complexity. The present study aims to understand and evaluate the characteristics and optimal trade-offs of various bottom cycles, including trans-critical, subcritical, and non-azeotropic mixed working fluids, by establishing 3E (energy, exergy and Exergoeconomic) models for sCO2/tCO2, sCO2/ORC, and sCO2/KC combined cycle systems. To analysis and enhance interpretability, statistical Global Sensitivity Analysis (GSA) alongside unsupervised learning-based Self-Organizing Map (SOM) techniques and parametric analysis are employed. These methods identify key parameters and discern system patterns. Subsequently, the Non-Dominated Sorting Whale Optimization Algorithm (NSWOA) and entropy weight-based Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) are applied to establish a Pareto-optimal decision space and identify compromised ideal design points for each combined cycle. The results quantitatively rank sensitivity for objective variables within the design space in different combined cycles and visually represent system nonlinearity and coupling relationships in 2D weight planes using SOM. The TOPSIS trade-off points based on the Pareto frontier for the three combined cycles are ηth = 45.08 %, cp,tot = 8.5199 $/GJ, ηth = 45.13 %, cp,tot = 8.3739 $/GJ, and ηth = 48.97 %, cp,tot = 8.4783 $/GJ, respectively. Integrating machine learning techniques provides a comprehensive understanding of system patterns, offering valuable insights for decision-makers in the design and optimization of combined cogeneration cycles.