We have proposed a frequency reshaping and compensation scheme for FTN (Faster-Than-Nyquist) CAP (Carrierless Amplitude and Phase modulation) signal based on machine learning for the first time in this paper. The cascaded post equalizer consists of 2 stages. The first stage digital signal processing (DSP) is a Deep Neural Network (DNN), then LMS post-equalization is conducted as the second stage offline DSP. Through this method, we successfully demonstrated a data rate of 1.12Gbit/s FTN CAP 9QAM modulation over 1meter free space transmission with bit error rate (BER) below 7% FEC threshold of 3.8×10-3. Compared to the traditional FTN CAP modulation without DNN, the proposed method would increase the data rate by approximately 100 Mbps, leading to an improvement of system capacity of 9.8%. The experimental results clearly validate the proposed cascaded two stage DNN-LMS (C2S DNNLMS) can be a promising solution for future high speed VLC system.