Abstract

According to the United Nations (n.d.), in 2018, around 23.5% of the urban population lived in slums or some other form of informal settlements; over 1 billion people worldwide. However, given the hindrances in surveying these population, such as lack of cooperation, scarcity of resources, and safety of our surveyors, many countries have moved to using big data to patch in the gap in their datasets. In an attempt to inform policies to give these populations their human rights and governmental benefits, such as public policies for education, health, social security and pensions, sanitation, or lack thereof, housing conditions and urban infrastructure, and the provision of essential services to homes and families, we used open-source satellite imagery, in combination with remote sensing machine learning and deep learning, to identify the informal settlements within the city of Rio de Janeiro. This is a preliminary study based on one 1km2 square, with the final goal being to fully classify the entirety of the city. We have run 8 different models, these being XGBoost, RandomForrest, LGBMClassifier, GradientBoost, Kmeans, GassianNB, LogisticRegression, and MLPClassifier. Two of these classifiers, the GradientBoost and XGBoost, have shown significant promise, being incredibly close by almost all metrics used. However, when it came to the IoU score, which we decided to be the best metric to judge the models by, the XGBoost has a 15% advantage on its closest contender and making it our best classifier currently, with a IoU Score of 75.6%.