one_class_SVM.py

def one_class_SVM_anomaly_detection(dataframe, columns_to_filter_by, outliers_fraction):
    """
    In this definition, time series anomalies are detected
    using a One Class SVM algorithm.
    Arguments:
        df: Pandas dataframe
        columns_to_filter_by: string, or list of strings. Name of the column(s) that 
        we want to use in the One Class SVM to detect time series anomalies
        outliers_fraction: float. Percentage of outliers allowed in the sequence.
    Outputs:
        df: Pandas dataframe with column for detected One Class SVM anomalies (True/False)
    """
    #Subset the dataframe by desired columns
    dataframe_filtered_columns=dataframe[columns_to_filter_by]
    #Transform the data using the fit_transform() function
    min_max_scaler = preprocessing.StandardScaler()
    np_scaled = min_max_scaler.fit_transform(dataframe_filtered_columns)
    scaled_dataframe = pd.DataFrame(np_scaled)
    #Remove any NaN's from the dataframe
    scaled_dataframe =scaled_dataframe.dropna()
    #Train the One Class SVM
    model = svm.OneClassSVM(nu=outliers_fraction, kernel="rbf", gamma=0.01)
    model.fit(scaled_dataframe)
    #Create a column for the anomaly
    one_class_svm_anomaly_column='One_Class_SVM_Anomaly'
    scaled_dataframe[one_class_svm_anomaly_column] = pd.Series(model.predict(
                                                        scaled_dataframe)).map({1: False, -1: True})
    dataframe[one_class_svm_anomaly_column] = scaled_dataframe[one_class_svm_anomaly_column]
    return dataframe

##EXECUTE IN MAIN BLOCK

#APPLY ONE CLASS SVM ALGORITHM TO DETECT ANOMALIES
#Create some columns to cluster against
#Create 6-value rolling average for gasoline prices
gasoline_price_df['Gasoline_Price_Rolling_Average_6_pt']=gasoline_price_df['Gasoline_Price'].rolling(window=6, center=True).mean()
#Calculate difference between gas price and rolling average value
gasoline_price_df['Gasoline_Price_Diff_From_Rolling_Avg']=gasoline_price_df['Gasoline_Price']-gasoline_price_df['Gasoline_Price_Rolling_Average_6_pt']

#Implement One Class SVM to time series
gasoline_price_df=one_class_SVM_anomaly_detection(dataframe=gasoline_price_df, 
                                    columns_to_filter_by=['Gasoline_Price_Diff_From_Rolling_Avg',
                                                          'Gasoline_Price'], 
                                    outliers_fraction=.1)
#Re-plot time series with color coding for anomaly column
scatterplot_with_color_coding(gasoline_price_df['Date'],
                                  gasoline_price_df['Gasoline_Price'], 
                                  gasoline_price_df['One_Class_SVM_Anomaly'],
                                  'Date', 
                                  'Gasoline Price (Dollars Per Gallon)', 
                                  'Gasoline Prices, Color-Coded on One Class SVM Anomalies')