To proactively enhance pedestrian safety using traffic conflict technique, it is necessary first to identify traffic conflicts among traffic interactions. Conflict indicators such as PET are used to quantify the proximity of traffic interaction to actual crash. These indicators are then compared to a threshold value to determine whether the interaction qualifies as a conflict. Three common approaches are used to determine threshold values for conflict indicators: select a predetermined threshold from the existing literature or use techniques such as crash conflict relationship or extreme value analysis. Selecting a predetermined threshold may not yield the best results, especially for heterogeneous traffic situations, and may require a more nuanced approach. The crash conflict relationship approach has the disadvantage of utilizing crash data, which sometimes can be unreliable, particularly in developing countries. The extreme value approach demands data to follow certain types of distribution, which may not be true in every case. This study addresses the limitations of existing methods by considering traffic conflicts as anomalies and employs Isolation Forest, a machine learning technique, to identify threshold value. An algorithm for automated extraction of PET values from video data is also used and validated. The proposed methodology was applied to an unsignalized intersection on an intercity highway with heterogeneous traffic. It established the PET threshold necessary to identify pedestrian-vehicle conflicts to be 2 s. The automated way of PET extraction has demonstrated excellent performance, as evidenced by its high R-squared value of 0.9756.