Comparison of Hidden Markov Model and KD-Tree in GPS Data-Based Map Matching Process

Data as a source of information in decision-making can be applied across various fields, such as government policy. Naturally, government policies have a significant impact because the entire society can feel their effects. Therefore, careful planning and structured implementation are required. Data can serve as one of the parameters for accurate planning. Policies can cover a wide range of scopes, including public mobility. An example of such a policy is the emergency health facilities built in the Java and Bali regions during the COVID-19 pandemic. Various types of data can be used in designing these policies, one of which is mobile positioning data (MPD). This data is advantageous because it can be collected quickly. However, MPD requires a complex process due to noise that can affect the information produced. One method to address this noise is map matching. The author is interested in comparing the accuracy levels of map matching using the Hidden Markov Model (HMM) algorithm and the KD-Tree algorithm. In this study, both algorithms process different categories of GPS data. Additionally, there is manipulation of the GPS data used, specifically random reduction. The parameter used to compare these algorithms is the similarity score obtained from the distance values in the Dynamic Time Warping (DTW) method. The test results show that the HMM algorithm has an accuracy level of about 80% for all tested data categories, while the KD-Tree algorithm has an accuracy level of about 55% for all tested data categories. It can be concluded that the HMM algorithm is superior in performing map matching for all data categories compared to the KD-Tree algorithm. Furthermore, it was found that a random reduction of GPS data can affect the resulting accuracy levels.