Visit and interact with the web app here!
To deploy a web application capable of providing predictions for apartment prices in Goiânia. The model used for prediction was Histogram Gradient Boosting Regressor. All you need to do is to access the web app, fill in the input data form and click "Predict!"!
The model used to predict the price of apartments in Goiânia was trained based on the main characteristics of the apartment (number of bedrooms, number of bathrooms, total area, floor, etc.) as well as additional characteristics of the ad, such as the main image and the full description of the advertiser. The mean percentage error of the model on the data separated for testing was 13.64%.
Relevant details
- Image features were extracted using ResNet-18 trained on the ImageNet-1k dataset;
- Features from the full description were extracted using TfIdfVectorizer;
- Missing values were processed natively by HistGradientBoostingRegressor;
- Hyperparameter tuning was done using Bayes optimization;
The static prediction model was trained using data collected from the internet (August, 2024), for educational purposes only. You should NOT use the model's predictions to make actual decisions or for commercial purposes. If you would like to better understand the steps that were taken to collect, clean and model the data, please consider studying the folders 1_data_extraction, 2_data_cleaning, 3_data_exploring, 4_img_feature_extraction and 5_modeling.
ApartmentPricesInGoiania by Carlos Eduardo Gonçalves de Oliveira is licensed under Creative Commons Attribution-NonCommercial 4.0 International
