[WEEK 3] Prediction Of Real Estate Price
Theme: Prediction Of Real Estate Price and Image Classification with Textual and Image Features
Team Members: Batuhan Ündar, Muhammed İkbal Arslan, Enes Koçak
Evocation
Our main goal in the project is to classify the pictures in the data we collect and estimate the house prices according to the parameters such as price, location, number of bathrooms, number of rooms, square meters, building age and luxury of the houses. We did data analysis last week. To determine the available data for our model, we investigated the effects of certain criteria on price.
This week we will talk about which methods we will concentrate on and how we get the efficiency from data we collect through these methods. First of all, we will start with “KERAS” to explain the libraries we use.
Keras: The Python Deep Learning library
Keras is a high-level neural networks API which was developed with a focus on enabling fast experimentation. Helps to create models quickly and easily. It supports both interoperable and repetitive networks and it works without problems on CPU and GPU.
Image Classification
For the image classification process, we examined the different CNN architectures such as AlexNet, GoogLeNet, VGG16, and VGG19. We had more work on the VGG architecture than the others.
VGG network is characterized by its simplicity, using only 3×3 convolutional layers stacked on top of each other in increasing depth. Reducing volume size is handled by max pooling. Two fully-connected layers, each with 4,096 nodes are then followed by a softmax classifier. We run this on the founded baseline code and we encountered some problems while running on CPU. So, I think we will get the best choice for the progress of the project as a result of trial and error.
Regression-Based on K-Nearest Neighbors
We use KNeighborsRegressor from sklearn.neighbors for seeing some regression results. After normalizing the positions as numbers, the lowest number is set to the position where the average price is the lowest, and the largest number is the position where the average price is the largest. In addition, when we regress according to the room, building age and number of bathrooms, for k = 13 value, we take the accuracy as %61 on train and %93 on the test.
After creating the model with using our dataset that we collect from real estate sites, we can change the dataset later for observation on calculations.
Support Vector Machines (SVM)
Support Vector Machine (SVM) is a supervised machine learning algorithm which can be used for both classification or regression challenges. It is one of the most effective and simple methods used in classification. For classification, it is possible to separate two groups by drawing a border between two groups in a plane. In order to do this, two boundary lines are drawn close to each other and parallel to each other and these boundary lines are brought together and common boundary line is produced.
While creating the SVM model, we use the number of square meters, rooms, bathrooms were used. The building age was not used as a parameter because it affected our results badly. Linear SVM results according to predicted values are above. Especially the data from “Ankara”. Additionally, we applied a linear regression model and the most effective feature on regression is surprisingly bathroom number…
This was our work and experiences this week. There are many methods we want to try. Neural network is one of them. I hope you found the article enjoyable. See you in the next weeks!
source: https://giphy.com/gifs/door-beware-salesmen-4rZrpWtowBzyw/links
