Abstract
This study illustrates a proof-of-concept model to improve solid waste management (SWM) services by analyzing people's behavior towards waste. A deep neural network model is implemented to detect and identify the specific types of events/activities in the proximity of the waste bin. This model consists of a three-dimensional convolutional neural network (3D CNN) and a long short-term memory (LSTM)-based recurrent neural network. The model was trained and tested over a handcrafted data set and achieved an average precision of 0.944-0.986. This precision is promising to support the implementation of the model on a large scale in the actual environment. The performance measures of all individual events indicate that the model successfully detected the individual events and has high precision for classifying them. The study also designed and built an experimental setup to record the data set, which comprises 3200 video files duration between 150-1200 s. Methodologically, the research is supported through a case study based on the recorded data set. In this case study, the frequencies of identified events/activities at a bin are plotted and thoroughly analyzed to determine people's behavior toward waste. This frequency analysis is used to determine the locations where one of the following actions is required to improve the SWM service: (i) people need to be educated about the consequences of waste scattering; (ii) bin capacity or waste collection schedules are required to change; (iii) both actions are required simultaneously; (iv) none of the actions are needed.