Advance House Price Prediction Report

Data Source: Kaggle

The dataset has record of 1460 houses which has 81 unique feature

Target: Based on these features predict the price

Data Preprocessing

Numerical Features: The dataset has 38 numerical feature

Problem: There are total 3 categories which has missing values. These values has relationship with target values.
Approach: Create 3 new columns, fill 0 where value is not missing and fill 1 where value is missing. Fill the Nan values with the median.
Status: This approach is effective.

Problem: When compare with Sale price Most of the numerical feature has outliers
Approach: Used IQR to find the Outliers and masked the outliers with lower limit and upper limit
Status: This approach is not effective. (NOTE: Model is not giving the generalized prediction)

Problem: The data is skewed
Approach: Log Transformation, Square root Transformation, Box-cox Transformation.
Status: Log Transformation works better than other 2. (Gives the highest accuracy )

Categorical Features: The dataset has 43 categorical feature

Problem: Categorical Feature
Approach: one hot encoding, Target Guided encoding (will groupby the categorical feature with the mean Sale price and give each category a rank based on the mean price).
Status: Target Guided encoding is Effective (Performed slightly better than one hot encoding)

Problem: Most of the Numerical feature has outliers.
Approach: Used IQR to detect the outliers and replace them with lower limit and upper limit.
Status: Not effective (Did Perform well on training and testing, Did not increased the Rank in Competition)

Problem: Every feature is not at the same scale.
Approach: Min Max Scaler, Standard Scaler, box-cox Transformation, Log Transformation.
Status: Log Transformation is effective. (Gives a slightly better results)
Scope of improving the approach is High

Linear Regression:
- Training Root Mean Squared Error – 0.122
- Test score – 0.132
- Score on kaggle – 0.1334
Support Vector Regression (kernel = polynomial, degree=4):
- Training Root Mean Squared Error – 0.116
- Test score – 0.124
- Score on kaggle – 0.1315
XGB Regressor:
- Training Root Mean Squared Error – 0.122
- Test score – 0.132
- Score on kaggle – 0.1438
- Scope to perform better – True
Random Forest Regressor:
- Training Root Mean Squared Error – 0.105
- Test score – 0.1470
- Score on kaggle – 0.1477
- Scope to perform better – True
ANN: (5 Hidden layer)
- Training Root Mean Squared Error – 0.099
- Test score – 0.015
- Score on kaggle – 0.136
Voting( Linear Regression, SVR and XGB)
- Training Root Mean Squared Error – 0.1083
- Test score – 0.1111
- Score on kaggle – 0.1243

Name		Name	Last commit message	Last commit date
Latest commit History 16 Commits
Data Preprocessing.ipynb		Data Preprocessing.ipynb
Deep Learning.ipynb		Deep Learning.ipynb
Feature Selection.ipynb		Feature Selection.ipynb
Model Selection.ipynb		Model Selection.ipynb
Model Tuning.ipynb		Model Tuning.ipynb
PCA.ipynb		PCA.ipynb
README.md		README.md
Testing Feature Transformation.ipynb		Testing Feature Transformation.ipynb
data_description.txt		data_description.txt
ft_test.csv		ft_test.csv
ft_train.csv		ft_train.csv
log_test.csv		log_test.csv
log_transformed.csv		log_transformed.csv
sample_submission.csv		sample_submission.csv
test.csv		test.csv
train.csv		train.csv