House price prediction using a hybrid GRU, MLP based on binary whale optimization algorithm and ant colony optimization for hyperparameter tuning

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Scientific Reports volume 16 , Article number: 23225 ( 2026 ) Cite this article

Accurate house price prediction is essential for real estate valuation, investment planning, and intelligent property decision-support systems. This study proposes an optimized hybrid deep learning framework that integrates a Gated Recurrent Unit and Multilayer Perceptron model with the Binary Whale Optimization Algorithm for feature selection and Ant Colony Optimization for hyperparameter tuning. The proposed framework was evaluated using a publicly available Kaggle house price regression dataset containing 500 housing records with structural, locational, and amenity-related attributes. The dataset was divided into training, validation, and testing subsets using a 70:20:10 ratio, and leakage-free normalization was applied using only the training data. Experimental results show that the proposed BWOA, ACO, GRU, MLP model outperformed standalone GRU, MLP, CNN, LSTM, and BiLSTM models. It achieved an MSE of 0.0146, MAE of 0.1051, RMSE of 0.1208, MAPE of 0.0112, MedAE of 0.0969, and an R 2 of 99.04%. These results demonstrate that combining feature selection, hyperparameter optimization, and hybrid neural regression improves prediction accuracy and model stability for house price estimation. The proposed framework provides a reliable data-driven approach for smart real estate valuation applications.

Housing price prediction is currently considered one of the most prominent topics within intelligent real estate analytics since the property evaluation is relevant to many processes, including investment, appraisal, tax assessments, urban development, and housing policies. Conventional housing price modeling techniques, including hedonic pricing and regression analysis, usually lack flexibility, rely on predefined assumptions, and are characterized by low sensitivity to nonlinear interactions between various factors such as structural characteristics, spatial effects, economic indicators, and environment-related attributes 1 , 2 . To address the mentioned limitations, the use of machine learning and deep learning methods was proposed in numerous academic studies, since they enable the learning of complex patterns from heterogeneous data and enhance the efficiency of automated valuation models. As an illustration, Park & Bae 3 presented machine learning algorithms for predicting housing prices and showed that data-driven methods were beneficial for real estate applications. Moreover, several other recent automated housing price valuation models confirmed the efficiency of machine learning and deep learning models.

Studies conducted have again shown that there is more to real estate price prediction than just using classical regression methods. With automated value estimation models, the use of ensembles, deep neural networks, gradient boosting, spatial information, description texts, and even explainable artificial intelligence can be utilized. Jafary et al. 1 developed automated valuation models for land that considered physical, geographic, socio-economic, environmental, legal, and planning information about the land. As shown by the study, it was possible for machine learning techniques like XGBoost to produce better results than classical methods. Baur et al. 4 , on the other hand, examined the use of machine learning for automated real estate valuation, considering property descriptions.

Despite this progress, predicting house prices still poses some challenges as there are many interacting variables that determine the cost of housing, such as house size, the number of rooms in a house, age of construction, quality of the location, presence of amenities, and proximity to cities. The relationships among these variables might be complex in nature and require more advanced approaches than traditional models to obtain highly accurate predictions. Recently, many studies on housing prices have revealed that machine learning algorithms can model nonlinear market dynamics and heterogeneities well, and explainable models have also been developed to understand how the location and properties influence prices 5 , 6 .

The other major challenge is model optimization. Machine learning and deep learning models are very dependent on hyperparameters like learning rate, the number of hidden layers, dropout ratio, batch size, and architecture depth. Hyperparameters selected wrongly will cause overfitting, instabilities in convergence, and poor generalization ability. Optimization algorithms are therefore critical in improving the accuracy of forecasts in hybrid models 7 . For example, Zhan et al. 8 suggested a hybrid Bayesian optimization algorithm using stacking, bagging, and transformer models for house price prediction. Metaheuristic optimization has proven effective in deep learning problems by discovering the best configurations compared to human intervention.

The deep learning architecture, including GRU, LSTM, CNN, and MLP, is extensively used in predicting tasks because of its ability to model nonlinear relations and interactions between different features. GRU networks are especially helpful because they offer a simplified recurrent architecture compared to LSTM, but still have the capacity to model sequence dependencies. Meanwhile, MLP is a powerful universal function approximation tool for tabular regression tasks that learns nonlinear functions mapping input variables to continuous output values. Recent hybrid models demonstrate that using GRU in combination with other neural network models improves prediction performance in nonlinear prediction tasks. At the same time, optimizing GRU networks leads to good results in real-world prediction tasks 9 , 10 .

However, the existing literature still presents several limitations. First, many studies rely mainly on standalone machine learning models, such as Random Forest, XGBoost, Support Vector Regression, or standard neural networks, without sufficiently investigating hybrid recurrent, feedforward architectures for house price regression. Second, although optimization algorithms have been applied in several forecasting domains, the integration of the Binary Whale Optimization Algorithm and Ant Colony Optimization for hyperparameter tuning in house price prediction remains underexplored. Third, many previous studies focus only on predictive accuracy but provide limited comparative analysis across deep learning baselines such as GRU, MLP, CNN, LSTM, and BiLSTM 11 , 12 . Fourth, some studies do not sufficiently analyze convergence behavior, training, validation stability, or error-based regression indicators such as MSE, MAE, RMSE, MAPE, MedAE, and R 2 together. These gaps motivate the development of a hybrid optimized deep learning model that can provide more accurate, stable, and generalizable house price predictions.

This study addresses the weaknesses by proposing a GRU, MLP hybrid architecture trained using the Binary Whale Optimization Algorithm for feature selection and Ant Colony Optimization to predict house prices. Although the housing dataset used in this study is tabular and does not contain natural temporal sequences, the GRU component is employed as a gated nonlinear feature-interaction encoder rather than as a conventional time-series forecasting model. After BWOA-based feature selection, each housing record is represented as a fixed ordered vector of normalized attributes, including structural, amenity-related, and locational variables. The GRU gates allow the model to selectively retain, suppress, and combine information across these attributes, thereby learning compact feature representations before the final regression stage. The MLP component is then used to map the learned representation to the continuous house price output. To assess the model’s efficacy in predicting house prices, an open-source housing price regression dataset from Kaggle with 500 instances is used. This data set incorporates numerous attributes related to structural and location characteristics of houses, such as square feet, number of bedrooms, number of bathrooms, number of floors, year built, presence of a garden, pool, garage size, location rating, proximity to downtown, and price. The data set is partitioned into training, validation, and test sets in the ratio of 70:20:10. The findings obtained reveal that the proposed GRU, MLP model performs better than other models, achieving MSE = 0.0146, MAPE = 0.0112, MedAE = 0.0969, MAE = 0.1051, RMSE = 0.1208, and R 2 = 99.04%.

The main contributions of this study are summarized as follows:

Proposing a hybrid deep learning model combining GRU and MLP for enhanced nonlinear regression in house price prediction.

Integrating BWOA for feature selection, effectively reducing redundancy and improving predictive performance.

Applying ACO for hyperparameter tuning to enhance convergence and model generalization.

Conducting a comprehensive comparative analysis with multiple deep learning baselines.

Demonstrating the effectiveness of combining feature selection and optimization techniques in a unified predictive framework.

The remainder of this paper is organized as follows. " Related works " section reviews related work on house price prediction and optimization techniques. " Materials and methods " section presents the proposed methodology, including BWOA-based feature selection and ACO-based hyperparameter tuning. " Experimental setup " section describes the experimental setup and evaluation metrics. " Results and discussion " section discusses the results and comparative analysis. Finally, " Conclusion and future work " section concludes the paper and outlines future research directions.

There have been considerable recent advances in house price prediction owing to the ever-increasing sophistication of the real estate market and the availability of data. Among traditional approaches for predicting real estate prices are hedonic regression models and index methods that have been popular over the years. However, their major limitation is that they are not efficient at modeling the complex relationship between various factors affecting house prices. Hence, there has been an exploration of numerous methods based on machine learning and deep learning algorithms to improve predictive performance and model robustness. This has also led to new research areas, including feature engineering and model optimization, in order to boost performance even further and increase interpretability. There still remain some problems with high-dimensional data, selecting appropriate features, avoiding overfitting, and defining model objectives consistent with real-world application scenarios. In this regard, this paper will review the most related studies on house price prediction utilizing machine learning models, hybrid modeling techniques, feature selection approaches, and optimization-based methods.

The research conducted by Adetunji et al. 13 . It is about the prediction of house prices while considering the drawbacks of conventional measures like the House Price Index (HPI). According to the research, the prediction of house prices should not rely solely on the identification of overall market trends, but it should also consider predicting the prices and/or the price ranges of properties. Predicting a variance or range would make sense and be applicable in practical situations, which made the problem a matter of classification. In order to mitigate the shortcomings of HPI, the researchers considered the inclusion of more attributes about the property and location of each property. The Random Forest algorithm