The demand for ultra-high data rates, ultra-low latency, and intelligent networking in 6 G massive-scale IoT environments necessitates efficient design strategies for multiple-input multiple-output (MIMO) antenna systems. This study presents a novel, compact dual-band 2 × 2 MIMO antenna optimized through machine learning (ML) techniques, addressing bandwidth limitations and design complexities while reducing computational overhead compared to conventional simulation methods for 6 G IoT applications. The proposed design features two orthogonally positioned T-slotted circular patch elements on a polyimide substrate, operating in the fundamental TM₁₀ mode to minimize electromagnetic leakage while enhancing radiation efficiency and directivity. Unlike traditional rectangular configurations, the slotted circular geometry delivers superior MIMO performance with broader bandwidth, enhanced isolation, reduced mutual coupling, and stable radiation characteristics in a compact footprint. Electromagnetic simulations were conducted using Computer Simulation Technology (CST) Studio Suite and cross-verified with Ansys HFSS, ensuring comprehensive validation. The antenna demonstrates exceptional dual-band performance with isolation of -32.2 dB, maximum bandwidth of 2.08 THz at 5 THz, and radiation efficiency of 87.43 %. Diversity metrics include an Envelope Correlation Coefficient (ECC) of 0.009 and Diversity Gain (DG) of 9.96, confirming excellent MIMO characteristics. Six ML regression models were employed for predictive performance optimization, with the Extra Tree regression model achieving superior accuracy exceeding 87 % for bandwidth prediction across all three frequency bands for both bandwidth and isolation parameters. The proposed circular MIMO antenna, validated through electromagnetic simulation and ML-based predictions, emerges as a promising candidate for terahertz 6 G IoT applications with excellent impedance matching, radiation performance, and diversity characteristics.