Outlier Detection

[SverreNystad]

def remove_outliers(df, value_col, period = None, robust = True):
    serie = df[value_col]
    res = STL(serie, period = period, robust = robust).fit()
    resid = res.resid
    q1 = resid.quantile(0.25)
    q3 = resid.quantile(0.75)
    iqr = q3 - q1
    lower = q1 - (3*iqr)
    upper = q3 + (3*iqr)

    anomalies = serie[(resid < lower) | (resid >= upper)]
    df = df.assign(anomaly = np.where(df[value_col].index.isin(anomalies.index), True, False))
    df["value_corrected"] = np.where(df["anomaly"] == True, np.NaN, df[value_col])
    df.interpolate(method = "linear", inplace=True)
    df["value_corrected"] = np.where(df["value_corrected"].isna(), df[value_col], df["value_corrected"])
    return df

[SverreNystad]

This made it worse