Adjusted KNN model in estimating user density in small areas with poor signal strength

Localized user density estimation is fundamental in many fields such as urban planning, traffic engineering, disease control, location based marketing and telecomm capacity planning. Modern mobility technologies provide the capability for measuring the localized user density dynamically and precisely. However, this is only limited to the areas that have good signal strength. It is a challenge to accurately estimate user density for areas with poor signal strength. However, user density can be estimated from other big data collected by telecommunication providers from different sources. This paper is a case study leveraging big data for developing a business solution. Exploratory Data Analysis (EDA) is applied to quantify the good signal vs bad signal, and a group of important variables that are highly related to user density are selected. An adjusted K-Nearest-Neighbor is applied to infer bad coverage user densities from the good coverage areas. Instead of predefining the K, different percentile measurements are provided to increase the robustness in business decision.