Existing explanation methods can highlight irrelevant patterns that coincidentally support predictions; TimePNS fixes this by measuring what actually breaks the model's decision when removed, giving more trustworthy explanations.
TimePNS is a framework that explains time-series classifier decisions by identifying which subsequences are truly necessary for predictions, not just sufficient. It uses counterfactual interventions to distinguish between spurious patterns that happen to support predictions and genuinely critical temporal factors, improving explanation faithfulness.