Prediction of solid household waste generation with machine learning in a rural area of Puno

• Autores: Cerna Cueva, Alberto Franco, Rosas Echevarría, Cesar Wilfredo, Perales Flores, Roberto Sixto, Ataucusi Flores, Pierina Lisbeth
• Formato: artículo
• Fecha de Publicación: 2022
• Institución: Universidad Nacional de Ingeniería
• Repositorio: Revistas - Universidad Nacional de Ingeniería
• Lenguaje: español
• OAI Identifier: oai:oai:revistas.uni.edu.pe:article/1378
• Enlace del recurso: https://revistas.uni.edu.pe/index.php/tecnia/article/view/1378
• Nivel de acceso: acceso abierto
• Materia: Desperdicios; Factor social; Algoritmos de machine learning; Gestión; Suburbios; Domicilio; Waste; Social factor; Machine learning algorithms; Management; Suburbs; Domicile

Solid waste management is one of the main environmental challenges in cities around the world due to factors such as population growth and consumption habits. One of the main tools for the design of waste management projects is the estimation of per capita generation, however, the traditional method to obtain this information demands a lot of effort and time, therefore this research proposes an alternative approach to estimate per capita generation based on socioeconomic factors. For this purpose, socioeconomic demographic information and information on the per capita generation of solid waste of 50 families was collected, subsequently the variables that have significant influence were determined from the correlation coefficient ρ of Spearman for numerical variables and an ANOVA for categorical variables with an acceptance threshold of 0.4 and 0.05 respectively. The selected variables were used to train the neural network, multiple linear regression, support vector machine, Gaussian process and random forest models, whose performances were R2 = 0.986, 0.982, 0.959, 0.837, 0.832; respectively. Cross validation and data partitioning were used for validation. The results indicate that the influential variables are per capita income, expenditure on supplies and products, family size and household services. It is concluded that the predictions of the models are reliable (RMSE from 8g to 27g) and from them projects can be designed.