Development of Rent House Price Prediction Service based on Machine Learning

Ⅰ. Introduction

The online market for a rental house is growing. One of the online marketplace platforms for the real estate rental sector is Airbnb, with over 150 million users, 650,000 property owners, and over 6 million properties registered in 2019. Airbnb serves as an alternative lodging place for tourists as well as an additional source of income for property owners [1].

Airbnb was founded in 2008 to share homes and experiences, and today has become one of the largest online accommodation booking platforms. Today, Airbnb offers more than 700 million properties in more than 220 countries in almost every country worldwide. Morgan Stanley predicts that Airbnb's users will continue to grow, and more people will use Airbnb instead of hotels. Hosts release free space online for guests looking for accommodation. The main reason for successful bookings is the right price, which is essential in balancing supply and demand in a two-sided market. Airbnb's current market share in China is 8.7%, much lower than other online platforms [2].

The Airbnb platform must help hosts set custom pricing effectively to expand market share. Thus, hosts and guests must be able to accommodate the price. Effective pricing offers help hosts earn more and entice guests to book through Airbnb rather than hotels or other platforms, which is an essential and challenging task for the Airbnb platform. Airbnb is an estate-sharing marketplace where property owners and tenants can publish properties online to allow guests to pay for accommodation [3]. As Airbnb is a platform that makes reservations in advance, unlike in-kind payments for hotels, it is essential to develop an intelligent pricing strategy that can accurately predict future house prices and offer reasonable prices to attract more tenants. Traditionally, Airbnb suggests that landlords first fix based on the price of a nearby or similar home, then increase or lower the cost if no one has booked at check-in or double the price during the holiday period. These strategies do not consider the flexible real market value of the home and can lead to difficulties in renting a home. Airbnb house prices can be affected by many factors [4]. Airbnb launched a price suggestion tool based on home properties in 2012 and improved its accuracy and flexibility in 2015 [5], but its price prediction model still needs improvement. Nowadays, hosts decorate their purchased or rented apartments to make money in the long run with Airbnb. Therefore, needs more suggestions for equipment, location, decorations, etc. [6]. The more users a site like Airbnb has, the more things to consider when pricing the rental house it offers.

Determining competitive rental rates is, thus, a complicated matter. This research addresses the challenge of creating a credible Rental House Price (RHP) prediction model using machine learning to assist the host in offering price evaluation and housing. This research's main contribution is to propose an approach for price prediction that uses multiple machine learning methods and has the advantage of learning patterns from the features of a dataset and making accurate predictions. We used three datasets: Airbnb's Seoul, Tokyo, and world. As for the machine learning method [7], experiments were performed through Linear Regression (LR), Decision Tree Regression (DTR), eXtreme Gradient Boosting Regression (XGBR), Random Forest Regression (RFR), Support Vector Regression (SVR), Multi-Layer Perceptron Regression (MLPR). The performance of each model was compared using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). In addition, via the commonly used machine learning methods, this work implements a web-based platform to predict the price quickly. The implemented web application can compare the information of different rental houses by predicting the price.

The rest of this research work is structured as follows. Section 2 discusses related work. Section 3 describes the proposed price prediction model. Section 4 outlines the experimental evaluations conducted on the model. Section 5 describes the implementation of the proposed model as a web. Finally, Section 6 concludes this paper.

Ⅱ. Related Work

Previous research work on price prediction models for Airbnb hosts was performed. A Gradient Boosting Machine (GBM) [8] was applied to predict the booking probability, which is the lodging demand curve. Then created a strategic model to provide a price offer. Finally, it provides accurate and personal advice, tailoring the proposal to the host's personal goals [9].

Researchers at Stanford University did a similar study. Pouya Rezazadeh Kalehbasti uses various methods such as machine learning and neutral language processing techniques [10] to develop price-prediction models. Pouya Rezazadeh Kalehbasti also uses several algorithms, including LR [11], SVR [12], K-means Clustering (KMC) [13], and Neural Networks (NNs) [14]. The main contribution of Pouya Rezazadeh Kalehbasti is the use of a tunable feature selection [15] technique that provides the 22 best features and a neutral network [16] to increase the model's accuracy. Pouya Rezazadeh Kalehbasti also added sentiment analysis to review customer reviews [17]. However, Pouya Rezazadeh Kalehbasti did not present a complete price prediction model.

Laura Lewis has studied pricing on London Airbnb listings, primarily through XGBoost and NNs. Laura Lewis also has over one year of host experience and performed exploratory data analysis [18].

In addition, some studies have worked on the influence characteristics of price. The features and information provided by the host are considered host-controlled variables. Features that can be controlled outside the host include variables managed externally. According to Brando MaNeil, both host-controlled and non-host-controlled variables are essential in determining Airbnb listing prices. Brando MaNeil also found that it significantly impacted listing price accuracy more when combined host-controlled with non-host-controlled variables. However, host-controlled variables have more significant implications for listing prices than non-host-controlled variables [19].

The prior studies show that renting an entire house has a higher average return than renting a room. According to Robbin Deboosere, the instant booking feature is another feature that affects average revenue. Moreover, commercial operators with multiple rentals can lower prices to get higher reviews and higher returns. These conclusions could potentially lead to the fact that the number of reviews can significantly impact booking probability and revenue. The development of nearby accommodation facilities also affects average income. As a result, hosts can earn higher revenue on average if they treat their house as a hotel rather than a temporary shared house [20].

Finally, Ang Zhu used regression and other machine learning models trained on New York City listings and attribution information from the Airbnb website, collected by Denis Gomonov and published on Kaggle. Robbin Deboosere's model analyzes the determinants of listing prices and predicts future prices for listings with publicly available information. Robbin Deboosere's model provides valuable information on the pricing strategies of hosts and other stakeholders, as well as insights into the entire hospitality industry in the context of the sharing economy [21].

Ⅲ. Proposed Approach

This study focused on predicting Airbnb prices by proposing a prediction model. The RHP prediction consists of data collection, pre-processing [22], and a prediction model. The proposed RHP prediction model, consisting of machine learning and deep learning [23] models, uses features from the pre-processed dataset obtained from online resources [24]. Using various machine learning methods, our model predicts the RHP by taking the relationships between parameters in the dataset. Figure 1 schematically depicts the proposed model.

Fig. 1.

Conceptual model fo the rental house price prediction method

Ⅳ. Experimentation

This section describes the experimental setup, dataset, hyperparameter tuning of the machine learning model, and the experiment's results. In this study, extensive experiments were performed to compare the performance of various well-known machine learning methods.

4-2 Experimental Setup

Deep learning and machine learning necessitate platforms with excellent computing performance. Table 1 shows the details of the experimental setup used in the experiments. TensorFlow, an open-source machine learning framework, was used to build the model. All experiments were performed on a personal computer with 62.7 GB RAM, an Intel Core i9-9900K CPU, an NVIDIA 1080 Ti GTX GPU, and running 64-bit Ubuntu 18.04 as the operating system. Python version 3.9.0 was used to train the proposed approach, and the TensorFlow 2.5.0 library and CUDA-Toolkit 11.2 were used on the GPU.

Table 1.

Experimental setup for performance

4-3 Hyperparameter Tuning for Machine Learning Models

The random search algorithm was used to adjust the hyperparameters of conventional machine learning methods. The random search algorithm exhaustively searches for a set of hyperparameters that optimize the model's results by taking a user-defined range of values as input. This process is performed during the training stage on the training set. The hyperparameters with the highest performance for each ML model were selected for the final models.

Table 2 describes the various optimizable parameters used during price prediction using traditional machine learning. In Table 2 the selected hyperparameter values of each algorithm can be seen after the random search algorithm that carefully determined the best value from the range of values as shown in the experimental source code. Table 2 is hyperparameter values for the Seoul, Tokyo, and World dataset, respectively.

Table 2.

Hyperparameters for the various machine learning models

4-4 Result

Our dataset was split into a training set and a test set to evaluate the prediction model's performance and avoid overfitting. This study compared commonly used regression evaluation metrics of machine learning tested in LR, DTR, XGBR, RFR, SVR, and MLPR. The mathematical equations for the regression evaluation metrics are defined as:

$\[ MAE=\frac{1}{n}\sum _{i=1}^n\left|{\widehat{y}}_i-y_i\right| \]$

(1)

MAE: Eq. (1) measures the average magnitude of the errors in a set of predictions without considering their direction. It’s the average over the test sample of the absolute differences between prediction and actual observation where all individual differences have equal weight.

$\[ MSE=\frac{1}{n}\sum _{i=1}^n{\left(y_i-{\widehat{y}}_i\right)}^2 \]$

(2)

MSE: Eq. (2) measures the amount of error in statistical models. MSE evaluates the average squared difference between the observed and predicted values. When a model has no error, the MSE equals zero.

$\[ RMSE=\sqrt[2]{\frac{1}{n}\sum _{i=1}^n{\left(y_i-{\widehat{y}}_i\right)}^2} \]$

(3)

RMSE: Eq. (3) is an evaluation that measures the error's average magnitude. Eq. (3) is the square root of the average square differences between prediction and actual observation. Since the errors are squared before averaged, the RMSE gives a relatively high weight to significant errors. This means the RMSE should be more useful when significant errors are undesirable.

Performance evaluation is suitable for selecting the best machine learning for prediction models after training. After regression analysis of price prediction performed in this work, tables 3, 4, and 5 show the performance results of all machine learning methods applied on the test set of the different collected datasets. We observed that RFR performed better than the other used machine learning methods. The proposed RFR approach displayed MAE, MSE, and RMSE of 14.10, 583.02, and 24.14, respectively as shown in Table 3. Table 4 shows the MAE, MSE, and RMSE of 17.90, 835.19, and 28.88, respectively. In addition, Table 5 shows the MAE, MSE, and RMSE of 19.80, 727.55, and 26.97, respectively. Therefore, the presented RFR results can be considered competitive compared to other applied prediction methods. RFR is the most appropriate algorithm for our work out of the machine learning methods assessed on the test set.

Table 3.

Performance comparison of machine learning models (Seoul dataset)

Table 4.

Performance comparison of machine learning models (Tokyo dataset)

Table 5.

Performance comparison of machine learning models (World dataset)

We split the dataset into the training of 80% and the test of 20% sets for our experiments, trained the model, and the hyperparameters were tuned using K-fold cross-validation on the validation set during the randomized search step. In our case, K = 5, training and validation were split into K subsets for validation and K-1 subsets for training during each fold. These results demonstrate that our model performed equally well on the training set and the unobserved data.

Features of each dataset were passed to the RFR model, and feature importance was calculated according to the function. Figure 2, 3, and 4 are graphs for arbitrary RFR. Since the RFR algorithm is random, different feature importance were obtained in each calculation. Figure 2 shows that the most important feature is bedrooms, with a score of 0.32.

Fig. 2.

Results of feature importance to RFR (Seoul dataset)

Fig. 3.

Results of feature importance to RFR (Tokyo dataset)

Fig. 4.

Results of feature importance to RFR (World dataset)

In Figure 2, the importance order of each feature is bedrooms > latitude > longitude > accommodates > bathrooms > room type entire home/apt > minstay > room type shared room > room type private room. Figure 3 shows that the most important feature is latitude, with a score of 0.28.

V. Implementation

This section discusses the model's actual web implementation and the test results obtained. The RHP prediction consists of data pre-processing and prediction. The data pre-processing phase performs data scaling. The prediction phase accurately predicts the price using the input data. During the deployment, a wireless local area network was set up to enable wireless communication between the web client and the server. In this paper, one PC was used as the server, and the other was used as the client side. The PC was running on Windows 10 64-bit and running on Chrome. Python's Flask web framework is used to build the server. Figure 5 provides an overview of the proposed implementation. First, the user inputs data about the rental house into the interface. Then data goes through pre-processing (data cleaning and normalization) to be further used by the RFR model to predict the owner's rental house price.

Fig. 5.

Web-based RHP prediction conceptual overview

Figures 6, 7, and 8 show the snapshot of the RHP application. Figure 6 shows a snapshot of the rental house information entered by the user. Users can predict RHP by entering values for Accommodation, bedrooms, bathrooms, latitude, longitude, minstay, and room type. Figure 7 is a snapshot of the results for Figure 6. First, the user can know the rental house information and the predicted value of RHP, and the user's surrounding rental house information and the surrounding RHP were displayed for comparison.

Fig. 6.

Snapshot of the user input form

Fig. 7.

Snapshot result of RHP prediction by location

Figure 8-① is the rental housing data in the dataset and Figure 8-② is the data entered by the user. Figure 7 displays the list of houses as markers on the map based on the user-selected preference. Figure 8 shows rental houses' information as a map marker, implemented to allow users to compare the rental house's prices in the same area. The user can insert room preference information (accommodates, bedrooms, bathrooms, latitude, longitude, and mainstay) on this page and select a room type.

Fig. 8.

Snapshot of the rental house’s information comparison from the selected location

Ⅵ. Conclusion

This paper proposes a web-based RHP prediction approach integrated with machine learning technology. The data pre-processing phase allows the system to make more accurate predictions. For the same purpose, this work compared the performance of the proposed algorithm with LR, DTR, XGBR, RFR, SVR, and MLPR. According to the experimental results on the Seoul Airbnb Dataset, the RFR model showed the best performance with MAE, MSE, and RMSE of 14.10, 583.02, and 24.14, respectively. Furthermore, in the Tokyo dataset, MAE, MSE, and RMSE were 17.90, 835.19, and 28.88, respectively, and in the world dataset, MAE, MSE, and RMSE were 19.80, 727.55, and 26.97, respectively. Data transformation and scaling were used to convert the data from text format to categorical format and perform normal distribution of the different selected features. The current implementation uses the owner's house data as input data to display price prediction results to the user. Users can also compare information from different houses on the map. Our dataset is constructed by removing all noise values from the Airbnb dataset. However, our model has limitations in not predicting accurate prices when noise values are input, which can be solved by adequately adding noise and proceeding with learning. Also, our model cannot determine the peak and off-peak times for forecast prices. Therefore, we plan to implement an updated version of the current prediction model to provide appropriate renting periods with lower prices to users.

Acknowledgments

This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government (MSIT) (No. 2019R1F1A1058394).

References

• Kokasih, M. F., et al, “Property Rental Price Prediction Using the Extreme Gradient Boosting Algorithm”, International Journal of Informatics and Information Systems, Vol. 3, No. 2, pp. 54-59, September 2020. [https://doi.org/10.47738/ijiis.v3i2.65]
• Hill, D, et al. (2020). “How much is your spare room worth?”, Ieee Spectrum, Vol. 52, No. 9, pp. 32-58, August 2015. [https://doi.org/10.1109/MSPEC.2015.7226609]
• Priambodo, F. N., & Sihabuddin, A., “An extreme learning machine model approach on airbnb base price prediction”, International Journal of Advanced Computer Science and Applications, Vol. 11, No. 11, pp. 179-185, 2020. [https://doi.org/10.14569/IJACSA.2020.0111123]
• Liu, Y., “Airbnb Pricing Based on Statistical Machine Learning Models”, in 2021 International Conference on Signal Processing and Machine Learning (CONF-SPML), pp. 54-59, November 2021. [https://doi.org/10.1109/CONF-SPML54095.2021.00042]
• Feng, L., “A scorecard breaking down everyone from Xiaozhu, Tujia to Airbnb”, EqualOcean, June 2019.
• Yang, S., “Learning-based airbnb price prediction model”, in 2021 2nd International Conference on E-Commerce and Internet Technology, pp. 283-288, March 2021. [https://doi.org/10.1109/ECIT52743.2021.00068]
• Kang, I. A., Ngnamsie Njimbouom, S., et al., “DCP: Prediction of Dental Caries Using Machine Learning in Personalized Medicine”, Applied Sciences, Vol. 12, No. 6, pp. 1-15, March 2022. [https://doi.org/10.3390/app12063043]
• Raghunath, K. K., Kumar, V. V., et al., “XGBoost Regression Classifier (XRC) Model for Cyber Attack Detection and Classification Using Inception V4”, Journal of Web Engineering, Vol. 21, No. 4, pp. 1295-1322, April 2022. [https://doi.org/10.13052/jwe1540-9589.21413]
• Ye, P., et al., “Customized regression model for airbnb dynamic pricing”, in Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, pp. 932-940, July 2018. [https://doi.org/10.1145/3219819.3219830]
• Luo, T. T., et al., “Crystal Engineering: Toward Intersecting Channels from a Neutral Network with a bcu‐Type Topology”, Angewandte Chemie, Vol. 117, No. 32, pp. 6217-6221, September 2005. [https://doi.org/10.1002/ange.200462674]
• Chattrairat, K., Wongseree, W., et al., “Comparisons of Machine Learning Methods of Statistical Downscaling Method: Case Studies of Daily Climate Anomalies in Thailand”, Journal of Web Engineering, Vol. 20, No. 5, pp. 1397-1424, July 2021. [https://doi.org/10.13052/jwe1540-9589.2057]
• Moon, C. B., Lee, J. Y., et al., (2020). “How to Retrieve Music using Mood Tags in a Folksonomy”, Journal of Web Engineering, Vol. 20, No. 8, pp. 2335-2360, November 2021. [https://doi.org/10.13052/jwe1540-9589.2086]
• Duo, J., Zhang, P., et al., “A K-means Text Clustering Algorithm Based on Subject Feature Vector”, Journal of Web Engineering, Vol. 20, No. 6, pp. 1935-1946, October 2021. [https://doi.org/10.13052/jwe1540-9589.20612]
• Zare, A., et al., “A Hybrid Recommendation System Based on the Supply Chain in Social Networks”, Journal of Web Engineering, Vol. 21, No. 3, pp. 633-660, February 2022. [https://doi.org/10.13052/jwe1540-9589.2133]
• Vandic, D., Frasincar, F., et al., “A framework for product description classification in e-commerce”, Journal of Web Engineering, Vol. 17, No. 1, pp. 001-027, October 2017.
• Guo, H., Li, et al., “Research on indoor wireless positioning precision optimization based on UWB”, Journal of Web Engineering, Vol. 19, No. 7, pp. 1017-1048, December 2020. [https://doi.org/10.13052/jwe1540-9589.19785]
• Rezazadeh Kalehbasti, et al., “Airbnb price prediction using machine learning and sentiment analysis”, in International Cross-Domain Conference for Machine Learning and Knowledge Extraction, pp. 173-184, August 2021. [https://doi.org/10.1007/978-3-030-84060-0_11]
• Lewis, L., “Predicting Airbnb prices with machine learning and deep learning”, Medium-Toward Data Science, 2019.
• McNeil, B. (2020). “Price Prediction in the Sharing Economy: A Case Study with Airbnb data”, University of New Hampshire, pp. 1-16, May 2020. https://scholars.unh.edu/honors/504/
• Deboosere, R., et al., “Location, location and professionalization: a multilevel hedonic analysis of Airbnb listing prices and revenue”, Regional Studies, Regional Science, Vol. 6, No. 1, pp. 143-156, April 2019. [https://doi.org/10.1080/21681376.2019.1592699]
• Zhu, A., et al., “Machine learning prediction of new york airbnb prices”, in 2020 Third International Conference on Artificial Intelligence for Industries (AI4I) , pp. 1-5, September 2020. [https://doi.org/10.1109/AI4I49448.2020.00007]
• Elbasani, E., & Kim, J. D., “LLAD: Life-log anomaly detection based on recurrent neural network LSTM”, Journal of Healthcare Engineering, Vol. 2021, No. 8829403, pp. 1-7, February 2021. [https://doi.org/10.1155/2021/8829403]
• Elbasani, E., Njimbouom, S. N., et al., “GCRNN: graph convolutional recurrent neural network for compound–protein interaction prediction”, BMC bioinformatics, Vol. 22, No. 5, pp. 1-14, January 2022. [https://doi.org/10.1186/s12859-022-04560-x]
• opendatasoft Available: https://public.opendatasoft.com/explore/
• Elbasani, E., & Kim, J. D., “AMR-CNN: Abstract Meaning Representation with Convolution Neural Network for Toxic Content Detection”, Journal of Web Engineering, Vol. 21, No. 3, pp. 677-692, February 2022. [https://doi.org/10.13052/jwe1540-9589.2135]
• Rodriguez-Galiano, V., et al., “Machine learning predictive models for mineral prospectivity: An evaluation of neural networks, random forest, regression trees and support vector machines”, Ore Geology Reviews, Vol. 71, pp. 804-818, December 2015. [https://doi.org/10.1016/j.oregeorev.2015.01.001]
• Fan, J., Wang, X., et al., “Comparison of Support Vector Machine and Extreme Gradient Boosting for predicting daily global solar radiation using temperature and precipitation in humid subtropical climates: A case study in China”, Energy conversion and management, Vol. 164, pp. 102-111, May 2018. [https://doi.org/10.1016/j.enconman.2018.02.087]
• Pekel, E., “Estimation of soil moisture using decision tree regression”, Theoretical and Applied Climatology, Vol. 139, No. 3, pp. 1111-1119, November 2020. [https://doi.org/10.1007/s00704-019-03048-8]
• Murtagh, F., “Multilayer perceptrons for classification and regression”, Neurocomputing, Vol. 2, No. 5-6, pp. 183-197, July 1991. [https://doi.org/10.1016/0925-2312(91)90023-5]
• Kavitha, S., et al., “A comparative analysis on linear regression and support vector regression”, in 2016 online international conference on green engineering and technologies (IC-GET), pp. 1-5, November 2016. [https://doi.org/10.1109/GET.2016.7916627]
• Rathore, S. S., & Kumar, S., “A decision tree regression based approach for the number of software faults prediction”, ACM SIGSOFT Software Engineering Notes, Vol. 41, No. 1, pp. 1-6, January 2016. [https://doi.org/10.1145/2853073.2853083]
• Zhou, J., Qiu, Y., et al., “Developing a hybrid model of Jaya algorithm-based extreme gradient boosting machine to estimate blast-induced ground vibrations”, International Journal of Rock Mechanics and Mining Sciences, Vol. 145, No. 104856, pp. 1-12, September 2021. [https://doi.org/10.1016/j.ijrmms.2021.104856]
• Adusumilli, S., Bhatt, D., et al. “A low-cost INS/GPS integration methodology based on random forest regression”, Expert Systems with Application, Vol. 40, No. 11, pp. 4653-4659, September 2013. [https://doi.org/10.1016/j.eswa.2013.02.002]
• Parbat, D., & Chakraborty, M., “A python based support vector regression model for prediction of COVID19 cases in India”, Chaos, Solitons & Fractals, Vol. 138, No. 109942, pp. 1-5, September 2020. [https://doi.org/10.1016/j.chaos.2020.109942]
• Yang, G. R., & Wang, X. J., “Artificial neural networks for neuroscientists: A primer”, Neuron, Vol. 107, No. 6, pp. 1048-1070, September 2020. [https://doi.org/10.1016/j.neuron.2020.09.005]
• Chen, T. L., et al., “A causal time-series model based on multilayer perceptron regression for forecasting taiwan stock index”, International Journal of Information Technology & Decision Making, Vol. 18, No. 6, pp. 1967-1987, November 2019. [https://doi.org/10.1142/S0219622019500421]