Under low-cost intensity modulation with direct detection (IM-DD) optical inter-connection scenarios, we demonstrate high-speed mode division multiplexing (MDM) transmission experiment over 1000 m length few-mode fiber (FMF) using two degenerate LP₁₁ modes enabled by multiple-input multiple-output (MIMO) equalizer based on a neural network (NN). With the help of the NN-based nonlinear MIMO equalizer, 2×100 Gbit/s MDM transmission with 30 G-class optical devices is achieved without the optical amplifier, and the single channel rate is approximately 100 Gbit/s. Four-pulse amplitude modulation (PAM-4) (2×50 Gbit/s) signals reached the 7% hard-decision forward error correction (HD-FEC) threshold with a sensitivity of approximately -10 dBm. Furthermore, a novel MIMO equalizer based on decision feedback neural networks (DFNN) is proposed from feature engineering, and the equalization performance is improved. This proposed method provides potential solutions for the future evolution of short-reach optical links.