In this paper, we proposed an analytical approach for latency-aware publish/subscribe systems on MANET. The proposed customisable model ensures the user a good quality of service requirements and combines both proactive and reactive statistical analyses. In fact, the reactive analysis, suitable for multimedia applications, aims to detect failures by approximating latency series with a Gumbel distribution. Besides, the proactive analysis, suitable for crisis management applications, forecasts failure occurrence relying on autoregression or autoregressive integrated moving average formulas. Finally, a hybrid analysis was proposed by dynamically switching from reactive to proactive forms of analysis whenever quality of service violations are noticed. The correlation method was also used to identify source failure causes once they were detected or predicted. The efficiency and accuracy of the proposed scheme were validated by performing simulations.