California Houses Prices Prediction

Brief Background

As we all know, California is one of the most famous state in the US. California has a lot of landmarks to visit, such as The famous Hollywood, Golden Gate Bridge, Silicon Valley, and many more. Therefore, there is always people from different state or abroad moving to California for Job, Education, Vacation, etc. With such high traffic from people coming to California, the demand of house also increase. especially people from abroad who knows very little of California, will have difficulty finding the best option for house. So I tried to create machine learning model to help end user to pick the best house based on few attributes

Link ML Deployment : https://house-price-predict.herokuapp.com/

Dataset

Calicornia House Price

Dataset Reference

Predictors

Parameter	Type	Description
longitude	float	Required. Longitude value for the block in California, USA
latitude	float	Required. Latitude value for the block in California, USA
housing_median_age	int	Required. Median age of the house in the block
total_rooms	int	Required. Count of the total number of rooms (excluding bedrooms) in all houses in the block
total_bedrooms	int	Required. Count of the total number of bedrooms in all houses in the block
population	int	Required. Count of the total number of population in the block
households	int	Required. Count of the total number of households in the block
median_income	float	Required. Median of the total household income of all the houses in the block
ocean_proximity	categorical	Required. Type of the landscape of the block [ 'NEAR BAY', '<1H OCEAN', 'INLAND', 'NEAR OCEAN', 'ISLAND' ]

target

Parameter	Type	Description
median_house_value	float	Required. Median of the household prices of all the houses in the block

Diagram Flow Preprocessing and EDA

Diagram Flow Feature Engineering

Diagram Flow Modeling and Deploying

Running API Call

POST API Call using Postman

• URL

https://house-price-predict.herokuapp.com

• Parameter

/predict_api

• Payload

  {
    "data":
    {
        "long": -122.23,	
        "lat": 37.88,
        "med_age": 41.0,
        "total_rooms": 880.0,
        "total_bedrooms": 129.0,
        "pop":	322.0,
        "hold":	126.0,
        "income": 8.3252,
        "ocean":"NEAR BAY"
    }
}

Response API Call

• success, 200 OK

POST API Call using curl

if you have no Postman installed on your machine, you can use curl to use predict_api. All you have to do just open your terminal and type in

curl --header "Content-Type: application/json" \
  --request POST \
  --data '{
    "data":
    {
        "long": -122.23,
        "lat": 37.88,
        "med_age": 41.0,
        "total_rooms": 880.0,
        "total_bedrooms": 129.0,
        "pop":  322.0,
        "hold": 126.0,
        "income": 8.3252,
        "ocean":"NEAR BAY"
    }
}' \
 https://house-price-predict.herokuapp.com/predict_api

Conclusion and Future Works

1. From the RMSE Score, XGBoost perform well for this dataset. although as we can see from comparison graph between actual data and trained model. trained model cannot predict actual data correctly.
2. For future works, train with the more recent dataset. To give more accurate prediction with recent situation, you have to use recent dataset with updated features.
3. Maybe you can improve more with Ensemble machine learning model to lower RMSE Score

Reference

1. Pacmann Course
2. Predicting House Prices with Machine Learning : https://towardsdatascience.com/predicting-house-prices-with-machine-learning-62d5bcd0d68f
3. House Price Prediction With Machine Learning in Python: https://medium.com/codex/house-price-prediction-with-machine-learning-in-python-cf9df744f7ff

Authors

• Widhar Dwiatmoko