Description:The human desire to improve their living standard is rising on daily basis, resulting in an increase in demand for housing and properties. However, the major challenge is that prospective buyers unable to evaluate the worth of a property precisely. This inability led to poor decisions on purchasing of property. Therefore, a forecasting system is highly required for getting the accurate selling prices, known as property sale price prediction. This system not only benefits buyers, also relax the sellers. A price value of any property varies depending on multiple features like its location, size, number of bedrooms and bathrooms, total residential area, garage capacity, age, roofing type, and other utilities. These features are independent features and have no impact on one another. The motive of the work is to forecast the sale price which is a dependent variable whose value influenced by the values of the independent variables. Deep learning algorithms enables machines to learn and execute tasks independently without the need for explicit programming instructions.
In this work, the first step is to analysing the dataset, which involves visualization of the data, finding missing values, and investigating the correlations between various factors to the property price. The following step, Feature Engineering, involves transforming raw data into qualities that significantly improve the effectiveness of deep learning model (Long short-term memory). Following that is Feature Selection process, where the most important features are selected based on the substantial impact on result prediction. Thereafter, the wavelet packet transform is applied to the time series input data to decompose into various subseries called high frequency and low frequency subseries. The subseries along with the other features allied to long short term memory deep learning model to forecast the target variable. However, the deployment of the wavelet packet transforms and long short-term memory is another crucial and important task for forecasting accuracy. For the wavelet transform packet, a comprehensive method is used to select the appropriate level and type of wavelet with frequency. On the other hand, the fine tuning of hyperparameters of long short-term memory is another important task. For this work, grid search optimization is used to select the appropriate hyperparameters. This fine tuning selects the optimal values of hyperparameters. To build the overall model the long short-term memory is trained to learn from data without explicit programming. Model deployment is the process of integration of a produced deep learning model into a real environment to make predictions based on past data. In the context of deep learning, prediction is the process of generating a result based on input data provided into a model, with prediction accuracy. This challenge is resolved using wavelet packet transform and Long Short-Term Memory, a deep learning approach for forecasting a continuous dependent variable from several independent factors.
, Claims:1. We claim that we dealt with housing price variations as a regression difficulty. Regression is a method for predicting the relationship between a desired dependent variable and a set of changing independent variables.
2. Our created model utilizes a effective signal processing technique along with a advanced Long Short Term Memory Deep learning model.
3. We claim that the developed model definitely useful for the buyers and sellers as well in terms for house selling price prediction. This approach will be revolutionary approach in the real estate market.
4. We claim that by using developed system, the investors can precisely estimate the true price of the property, even for the unknown city.
5. To increase the exposure of the commercial property, the work offers an online platform.