Location-Centered House Price Prediction: A Multi-Task Learning Approach

Accurate house prediction is of great significance to various real estate stakeholders such as owners, buyers, and investors. We propose a location-centered framework that differs from existing work in terms data profiling model. Regarding profiling, we make an important observation follows – besides the in-house features floor area, location plays critical role price prediction. Unfortunately, either overlooked it or had coarse grained measurement locations. Thereby, define capture fine-grained profile powered by diverse range sources, including transportation profile, education suburb based on census data, facility profile. choice model, observe variety approaches consider entire for modeling, split model each partition independently. However, modeling ignores relatedness among partitions, all scenarios, there may not be sufficient training samples per latter approach. address this problem conducting careful study exploiting Multi-Task Learning (MTL) Specifically, map strategies splitting ways tasks are defined MTL, select specific MTL-based methods with different regularization exploit tasks. Based real-world transaction collected Melbourne, Australia, design extensive experimental evaluations, results indicate significant superiority over state-of-the-art approaches. Meanwhile, conduct in-depth analysis impact task definitions method selections MTL performance, demonstrate performance far exceeds selections.