In this paper, belief propagation (BP) multiple-input multiple-output (MIMO) detection with minimum mean square error (MMSE) pre-cancellation matrix switching in overloaded MIMO systems is proposed. Loops in a factor graph degrades the demodulation performance of the BP scheme. Therefore, several schemes that suppresses the negative impact of loopy BP have been proposed. Among them, the application of MMSE pre-cancellation prior to BP has been proposed. In the scheme, the effects of loops in the factor graph is suppressed by MMSE pre-cancellation. However, many loops with large coefficients remain in the factor graph even after MMSE pre-cancellation. Therefore, the proposed scheme applies MMSE pre-cancellation matrix switching every after a preset number of BP iterations in order to suppress the occurrence of large beliefs in the loops. Numerical results obtained through computer simulation show that the proposed scheme achieves better BER performance than BP with MMSE pre-cancellation without matrix switching. The proposed scheme outperforms the conventional without switching by 0.7 dB at BER= 10-3.