Boosting Algorithms Notebook to Predict Flood Part 1

				
					# Collect unique values for all feature columns in the training and testing datasets
unique_vals = []
for df in [train, test]:
    for col in initial_features:
        unique_vals += list(df[col].unique())
unique_vals = list(set(unique_vals))

from sklearn.preprocessing import StandardScaler

# Generate new features
def getFeats(df):
    # Standardize numerical feature columns
    # Ensuring different scale features have the same scale, which helps the model's learning and convergence.
    scaler = StandardScaler()
    
    # Create interaction features
    # Climate and human activity interaction features: Reflecting the combined impact of climate change intensity and human activities such as deforestation, urbanization, etc.
    df['ClimateAnthropogenicInteraction'] = (df['MonsoonIntensity'] + df['ClimateChange']) * \
                                            (df['Deforestation'] + df['Urbanization'] + df['AgriculturalPractices'] + df['Encroachments'])
    
    # Infrastructure and disaster management interaction features: Examining the interaction between infrastructure quality and inadequate disaster preparedness.
    df['InfrastructurePreventionInteraction'] = (df['DamsQuality'] + df['DrainageSystems'] + df['DeterioratingInfrastructure']) * \
                                                (df['RiverManagement'] + df['IneffectiveDisasterPreparedness'] + df['InadequatePlanning'])
    
    # Create descriptive statistical features
    # Basic statistics: Providing the model with simple statistical descriptions of the raw data.
    df['sum'] = df[initial_features].sum(axis=1)   # Sum of features
    df['std'] = df[initial_features].std(axis=1)   # Standard deviation
    df['mean'] = df[initial_features].mean(axis=1) # Mean
    df['max'] = df[initial_features].max(axis=1)   # Maximum value
    df['min'] = df[initial_features].min(axis=1)   # Minimum value
    df['mode'] = df[initial_features].mode(axis=1)[0]  # Mode
    df['median'] = df[initial_features].median(axis=1) # Median
    df['q_25th'] = df[initial_features].quantile(0.25, axis=1) # 25th percentile
    df['q_75th'] = df[initial_features].quantile(0.75, axis=1) # 75th percentile
    df['skew'] = df[initial_features].skew(axis=1) # Skewness
    df['kurt'] = df[initial_features].kurt(axis=1) # Kurtosis
    
    # Range feature: Check if the sum of features falls within a specific range.
    df['sum_72_76'] = df['sum'].isin(np.arange(72, 76))
    
    # Quantile features: Providing the model with more detailed data distribution information.
    for i in range(10, 100, 10):
        df[f'{i}th'] = df[initial_features].quantile(i / 100, axis=1)
    
    # Other mathematical features
    df['harmonic'] = len(initial_features) / df[initial_features].apply(lambda x: (1/x).mean(), axis=1) # Harmonic mean
    df['geometric'] = df[initial_features].apply(lambda x: x.prod()**(1/len(x)), axis=1) # Geometric mean
    df['zscore'] = df[initial_features].apply(lambda x: (x - x.mean()) / x.std(), axis=1).mean(axis=1) # Z-score mean
    df['cv'] = df[initial_features].std(axis=1) / df[initial_features].mean(axis=1) # Coefficient of variation
    df['Skewness_75'] = (df[initial_features].quantile(0.75, axis=1) - df[initial_features].mean(axis=1)) / df[initial_features].std(axis=1) # 75th percentile skewness
    df['Skewness_25'] = (df[initial_features].quantile(0.25, axis=1) - df[initial_features].mean(axis=1)) / df[initial_features].std(axis=1) # 25th percentile skewness
    df['2ndMoment'] = df[initial_features].apply(lambda x: (x**2).mean(), axis=1) # Second moment
    df['3rdMoment'] = df[initial_features].apply(lambda x: (x**3).mean(), axis=1) # Third moment
    df['entropy'] = df[initial_features].apply(lambda x: -1*(x*np.log(x)).sum(), axis=1) # Entropy
    
    # Unique value count features: Counting the occurrences of each unique value in the dataset, helps understand data discreteness.
    for v in unique_vals:
        df['cnt_{}'.format(v)] = (df[initial_features] == v).sum(axis=1)
    
    df[initial_features] = scaler.fit_transform(df[initial_features])
    
    return df
    
# Add new features to the training and testing datasets
train['type'] = 0
test['type'] = 1
all_data = pd.concat([train, test], axis=0)
all_data = getFeats(all_data)
all_data

# Split training and testing datasets
train = all_data[all_data['type']==0]
test = all_data[all_data['type']==1]

# Split the features and target variable
X_train = train.drop(['FloodProbability', 'type'], axis=1)
y_train = train['FloodProbability']
X_test = test.drop(['FloodProbability', 'type'], axis=1)