You will be evaluated on classification accuracy: that is the proportion of correct class labels. Although each classification is effectively a multi-label classification (e.g. +fridge+washing_machine) there are no points for getting one of the labels correct. Each combination is treated as an individual class. 

The public leaderboard shows the classification accuracy of submissions on a subset of the data (30%). The actual results, evaluated on the remaining 70%, will be different and are only released at the end of the competition.