A DL-Based Garbage Management System

Abstract

Automated machine learning is presented as a viable solution to a real-world issue in Smart Waste Management. This article focuses on the difficulty of detecting an emptying of a recycling container using sensor readings, or binary classification. In a realistic environment, where the vast majority of the occurrences were not true emptyings, a wide variety of data-driven approaches to the issue were explored. Both the current model, which was manually developed, and a modified version of it, as well as more traditional machine learning techniques, were among the strategies explored. With the use of machine learning, the current manually constructed model's classification accuracy and recall were increased to 99.1% and 98.2%, respectively. Based on the filling level at certain intervals, a collection of features were generated and fed into a Random Forest classifier to arrive at this answer. Finally, the top performing solution enhanced the quality of predictions for emptying time of recycling containers in comparison to the baseline existing manually created model.