Machine learning for semi-automated land parcel plotting

The Indonesian cadastral quality improvement process, established in 2018, faces huge challenges due to its huge land area, diverse natural and social conditions, and historical mapping practices. With 15 million land parcels yet to be plotted accurately, in this work we propose a solution based on machine learning to speed up the process of plotting parcels documented on old land certificates and maps. Our approach is a heuristic-driven geospatial data matching procedure that semi-automatically searches for possible locations for plotting land parcels. The heuristic aspect of the approach is its basis in the manual plotting method in use presently in Indonesia. We identified eight causes of unplotted parcels and for two of these causes we have implemented an optimization algorithm model defined over five geometric parcel attributes. The model is then optimized using the RCGA algorithm. The model performed with recall rates of 88.3% and 57% on test data and real data respectively, and precision rates of 98.8% and 91% also respectively for test data and real data. After the incorporation of text information in old cadastral maps, the recall value for real data improved to 91%.