CASE

The project aims to predict the price of each house in Iowa. There are 79 independent variables in the data set.

• Kaggle notebook: Here

WHY

This is an individual project from Kaggle that I worked on to practice my data science skills.

I built advanced regression models to predict the final price of each house with 79 independent variables.

The whole code & work can be found from Kaggle notebook : Here

WORKFLOW

• Collection of data from Kaggle Titanic dataset consisting of Train & Test data
• Exploratory data analysis
• Data preprocessing
• Model & Prediction
• Evaluation
• Submission

EXPLORATORY DATA ANALYSIS

This is the screenshot of matrix that shows the correlation among independent variables. The strong correlation exhibites variable ‘GarageYrBlt’ & ‘YearBuilt’, ‘TotalBsmtSF’ & ‘1stFlrSF’, ‘TotRmsAbvGrd’ & ‘GrLivArea’, ‘GarageArea’ & ‘GarageCars’. 

DATA PREPROCESSING

• Analysis of Dependent variable
• Dealing with outliers
• Dealing with missing value
• Normalisation of dependent variable
• Feature creation
  • Create new feature
• Feature transformations
  • Conversion of data types
  • Normalisation of numierical features
  • Encoding categorical features

skew_features = all_X[Numerical].apply(lambda x: skew(x)).sort_values(ascending=False)

high_skew = skew_features[skew_features > 0.5]
skew_index = high_skew.index

print("There are {} numerical features with Skew > 0.5 :".format(high_skew.shape[0]))

skewness = pd.DataFrame({'Skew' :high_skew})
skew_features.head(10)

MODEL & PREDICTION

• Cross validation
• Ridge
• Lasso
• Elastic net
• Random forest
• GBR
• XGBoost
• LightGBM

# Hyperparameter Tuning 

lgbm = LGBMRegressor(boosting_type='gbdt',
                     objective='regression',
                     n_estimators=5000,
                     num_leaves=4, 
                     learning_rate=0.01, 
                     max_depth=2,
                     lambda_l1=0.0001, 
                     lambda_l2=0.001, 
                     min_child_samples=10, 
                     feature_fraction=0.4, 
                    verbose= -1)

# Scoring 

score = cv_rmse(lgbm)
print("lgbm score: {:.4f} ({:.4f})".format(score.mean(), score.std()))
scores['lgbm'] = (score.mean(), score.std()) 

# Fitting a model 

lgbm_model = lgbm.fit(train_X, train_y)
print('lgbm')

# Training a model 

lgbm_train_pred = lgbm_model.predict(train_X)
lgbm_pred = np.expm1(lgbm_model.predict(test_X))
print(rmsle(train_y, lgbm_train_pred))

EVALUATION

Based on scores above, we can sort the scores of all the models.

We decide to use

The whole code & work can be found from Kaggle notebook : Here

💼 Looking for a data analyst?

🚀 View my portfolio

I specialise in turning raw data into actionable insights using SQL, Python, and Power BI.