In this article, we propose an adaptive multiple stage K-best successive interference cancellation (AMS-KSIC) algorithm for symbol vector detection in multiple-input multiple-output systems. The proposed algorithm employs multiple successive interference cancellation (SIC) stages in parallel, where the number of stages depends on the number of positions at which the minimum mean squared error (MMSE) estimate of the received vector and the SIC solution differ, and each stage is initialized with the partial MMSE estimate of the received vector. In every stage, K-best solutions are generated by using the minimum Euclidean distance criteria. Furthermore, to reduce error propagation, we use two different ordering strategies namely, signal to noise ratio and log-likelihood ratio based orderings. The best solution among all the generated solutions is selected by using maximum likelihood (ML) cost metric. Multiple stages along with K-best solutions in every stage achieves a higher detection diversity, and hence, yield a better performance in terms of bit error rate (BER). From simulations, we observe that the proposed AMS-KSIC algorithm performs better than the MMSE and the SIC based detection schemes, and achieves a near ML performance. Further, the BER performance of the proposed algorithm improves with increase in the number of antennas and shifts towards single-input single-output additive white Gaussian noise performance. In addition, we also check and validate robustness of the proposed algorithm by simulating the BER performance under channel estimation errors.