Active learning for automatic classification of complaints about municipal waste management

Information flow is an important issue in the area of waste management. There is a need for a fast response to reported problems. Therefore we investigated the classification process of Polish wasterelated complaints sent by Wrocław’s residents. It has been noticed that residents, mostly without expert knowledge of waste management, incorrectly classify the observed problems. In response to the observed unacceptable classification accuracy, we introduced a multi-class machine learning classification. Machine learning is widely used in waste management issues like predicting waste generation or different waste fractions identification for automated sorting. However, based on the literature review, it can be stated that there is a lack of solutions in machine learning-based text classification regarding waste management. Ten chosen classifiers were used to classify considered complaints into defined categories automatically. Additionally, we incorporated the active learning approach to reduce experts' effort involved in the labeling process, which is necessary when having an unlabeled dataset. The results confirm the possibility of applying machine learning algorithms to waste-related Polish complaints.