House Rent Estimation in Dhaka City by Multi Layer Perceptions Neural Network

Samsuddin Ahmed, Md. Mahbubur Rahman, Sabirah IslamInternational Journal of u- and e- Service Science and Technology 2014 · DOI

Abstract

This paper attempts to build an artificial neural network that can estimate the median price of a home in a neighborhood described by forty demographic attributes in areas of Dhaka City namely Dhanmondi, Baridhara, Gulshan, Mirpur, Uttara and Old Dhaka. A structured questionnaire was used to collect the relevant data and the housing data sets was used to develop constant quality price indices using traditional econometric techniques and using neural networks incorporating genetic algorithms. Factors including house size, house age, house type, number of bedrooms, number of bathrooms, number of garages, amenities around the house and geographical location are considered. The analysis indicates that neural networks act as real alternative to the econometric methods. In this paper thirteen conditional attributes have been considered to estimate the house rent. This paper provides some indicative policy guidelines to handle the house rent problem in the Dhaka City and suggests that a rent controller should be appointed for each ward and maximum rents for particular areas should be gazette and the rents should be paid through banks.

Introduction

Dhaka, one of the world's most densely populated megacities, faces a significant housing crisis driven by massive in-migration. This demand has caused house rents to skyrocket, creating a "hostage condition" where many tenants spend over 65% of their monthly income on housing. A primary driver of this instability is the lack of a standardized pricing mechanism, allowing landlords to increase rents arbitrarily. This study proposes a data-driven approach using Multi-Layer Perceptron (MLP) neural networks to establish a reliable tool for rent estimation and fair market valuation.

Methodology

The research utilizes a Multi-Layer Perceptron (MLP), a feed-forward artificial neural network architecture. While the broader study identified 40 potential demographic and physical attributes, the final predictive model focuses on 13 key conditional attributes as inputs.

Technical Configuration

  • Architecture: A two-layer feed-forward network comprising an input layer, a hidden layer (configured with 10 neurons), and an output layer.
  • Training Algorithm: The model employs the Levenberg-Marquardt backpropagation algorithm, optimized for function fitting in small-to-medium datasets.
  • Activation Functions: Sigmoid neurons are utilized in the hidden layer to capture non-linearities, while linear output neurons allow for continuous rent value predictions.
  • Validation Strategy: The dataset was partitioned into training (70%), validation (15%), and testing (15%) subsets to ensure robust generalization.

Dataset and Geographic Variance

The study is based on a primary dataset collected through a field survey of 500 flats across five major residential areas: Dhanmondi, Baridhara, Gulshan, Mirpur, and Uttara. The survey captured a wide range of property sizes (750 to 3,000+ sq. ft.) and attributes such as building age, room counts, and proximity to essential services.

Table 1: Frequency and average rent per square foot (Tk.) across different areas of Dhaka.

The data reveals significant geographic disparities. Gulshan recorded the highest average rent at approximately 36.26 Tk/sq.ft, whereas Mirpur emerged as the most affordable area at 23.67 Tk/sq.ft. Regression analysis confirms that property size remains the most dominant predictor of rent variation in the local market.

Key Results and Performance

The neural network demonstrated high predictive accuracy, achieving a regression value of R = 0.92516. This indicates that the model can explain roughly 92.5% of the variation in house rents based on the input features.

Figure 1: Training performance plot showing Mean Squared Error (MSE) convergence, with best validation performance at epoch 10.

As shown in the performance plot, the model reached its best validation performance with a Mean Squared Error (MSE) of 11.2191 at epoch 10. This rapid convergence suggests that the MLP architecture is highly effective for this specific regression task.

Conclusion and Policy Implications

The findings suggest that neural networks are a superior alternative to traditional hedonic econometric models for real estate valuation due to their ability to model complex, non-linear relationships. Beyond technical validation, the study advocates for urban governance reform. The authors propose that these models be used by government-appointed "rent controllers" to gazette maximum allowable rents per ward, ensuring a more equitable and regulated housing market in Dhaka City.

Future work should aim to incorporate temporal dynamics (the "time effect" on rent) and macro-economic factors like interest rates to further refine the model's predictive power.

Summary updated

Abstract

This paper attempts to build an artificial neural network that can estimate the median price of a home in a neighborhood described by forty demographic attributes in areas of Dhaka City namely Dhanmondi, Baridhara, Gulshan, Mirpur, Uttara and Old Dhaka. A structured questionnaire was used to collect the relevant data and the housing data sets was used to develop constant quality price indices using traditional econometric techniques and using neural networks incorporating genetic algorithms. Factors including house size, house age, house type, number of bedrooms, number of bathrooms, number of garages, amenities around the house and geographical location are considered. The analysis indicates that neural networks act as real alternative to the econometric methods. In this paper thirteen conditional attributes have been considered to estimate the house rent. This paper provides some indicative policy guidelines to handle the house rent problem in the Dhaka City and suggests that a rent controller should be appointed for each ward and maximum rents for particular areas should be gazette and the rents should be paid through banks.
BibTeX
@article{ahmed2014house,
  author = {Samsuddin Ahmed and Md. Mahbubur Rahman and Sabirah Islam},
  title = {{House Rent Estimation in Dhaka City by Multi Layer Perceptions Neural Network}},
  journal = {International Journal of u- and e- Service Science and Technology},
  year = {2014},
  doi = {10.14257/ijunesst.2014.7.4.26},
}
Figure