For decades of battling poverty, Malaysia’s poverty rates significantly reduced from 49% in the year 1970 to 0.4% in the year 2016. Despite the government giving the very best efforts in creating poverty alleviation programs, we still unable to completely wipe out poverty in Malaysia. According to (Tingzon et al., 2019), one of the utmost challenges to battle with poverty is the lack of well-founded socioeconomic data. These data usually are expensive, labor-intensive, and time consuming to collect. For example, the Philippines required a huge cost of up to 1.5 million USD to conduct such a survey (Tingzon et al., 2019). In Malaysia, the census survey will only be conducted twice every five years. The last census data was conducted in the year of 2015. Advancements in computer vision research coupled with the widely available sources of Geo-spatial data in recent years enable us to estimate socioeconomic indicators (Jean et al., 2016). Estimating socioeconomic indicators from nightlight intensity and day-time satellite images has been used to predict poverty at a larger scale in fully developed countries (Elvidge et al., 2009). In order to ease the identification of the poverty, we hereby propose a method of combining both machine learning and satellite images as a quick, cost saving and highly extensible approach for detecting poverty.

Measuring poverty is very important for a local policymaker to target in a place that needs aids by any term with the most effectiveness. However, it may be difficult, expensive, and labor-intensive to collect detailed data from every household in Malaysia. Hence, data gaps is happening with every five years interval. Because of that, policymakers are often making decisions based on old data that collect back years ago. Making decisions based on outdated data by policymakers can create impactful problems in the future that may worsen the situations instead of solving it. Currently, in Malaysia, the government are still using manual surveys method to gather economic measurements and take actions. Applying these surveys is not simple and very costly and labor-intensive. With the advancement of machine learning technology, it is now possible to detect poverty in Malaysia with the help of intensity of night light and also day time satellite image.

Objective
1. To collect a satellite dataset of Malaysia to be used for poverty detection.
2. To train a model that can detect poverty regions in North Malaysia satellite images.
3. To provide visualization of the detected regions poverty.

Scope
1. The results of poverty detection is scoped as multiclass classification of four
levels of poverty.
2. The method will be trained on day-time satellite image.
3. The target regions are NorthernMalaysia; Penang, Kelantan, Perak, Perlis, Kedah