Package funModeling: data cleaning, importance variable analysis and model perfomance

╰半橙微兮° 2022-08-20 06:29 258阅读 0赞

(This article was first published on  **[R - Data Science Heroes Blog][]**, and kindly contributed to  [R-bloggers)][R-bloggers]

![Fancy miniature][]

Hi there ![:)][simple-smile.png_zoom_3_w_456]

This new package –`install.packages("funModeling")`– tries to cover with simple concepts common tasks in data science. Written like a *short tutorial*, its focus is on data interpretation and analysis.

Below, you’ll find a copy-paste from the package vignette, (so you can drink a good coffee while you read it… )

--------------------

#### Introduction ####

This package covers common aspects in predictive modeling:

1.  **Data Cleaning**
2.  **Variable importance analysis**
3.  **Assessing model performance**

Main purpose of this package is to teach some predictive modeling using a practical toolbox of functions and concepts, to people who is starting in data science, small data and big data. With **special focus** on results and analysis understanding.

#### Part 1: Data cleaning ####

**Overview**: Quantity of zeros, NA, unique values; as well as the data type may lead to a good or bad model. Here an approach to cover the very first step in data modeling.

## Loading needed libraries
    library(funModeling)  
    data(heart_disease)

###### Checking NA, zeros, data type and unique values ######

my_data_status=df_status(heart_disease)

![Variable description with df\_status][Variable description with df_status]

*  `q_zeros`: quantity of zeros (`p_zeros`: in percentage)
 *  `q_na`: quantity of NA (`p_na`: in percentage)
 *  `type`: factor or numeric
 *  `unique`: quantity of unique values

###### Why are these metrics important? ######

*  **Zeros**: Variables with **lots of zeros** may be not useful for modeling, and in some cases it may dramatically bias the model.
 *  **NA**: Several models automatically exclude rows with NA (**random forest**, for example). As a result, the final model can be biased due to several missing rows because of only one variable. For example, if the data contains only one out of 100 variables with 90% of NAs, the model will be training with only 10% of original rows.
 *  **Type**: Some variables are encoded as numbers, but they are codes or categories, and the models **don’t handle them** in the same way.
 *  **Unique**: Factor/categorical variables with a high number of different values (~30), tend to do overfitting if categories have low representative, (**decision tree**, for example).

###### Filtering unwanted cases ######

Function `df_status` takes a data frame and returns a the status table to quickly remove unwanted cases.

**Removing variables with high number of NA/zeros**

# Removing variables with 60% of zero values
    vars_to_remove=subset(my_data_status, my_data_status$p_zeros > 60)  
    vars_to_remove["variable"]

![Variable description with df\_status][Variable description with df_status 1]

## Keeping all except vars_to_remove 
    heart_disease_2=heart_disease[, !(names(heart_disease) %in% vars_to_remove[,"variable"])]

**Ordering data by percentage of zeros**

my_data_status[order(-my_data_status$p_zeros),]

![Variable description with df\_status][Variable description with df_status 2]

#### Part 2: Variable importance with cross\_plot ####

*  **Overview**:
    
     *  **Analysis purpose**: To identify if the input variable is a good/bad predictor through visual analysis.
     *  **General purpose**: To explain the decision of including -or not- a variable to a model to a *non-analyst* person.

**Constraint:** Target variable must have only 2 values. If it has `NA`values, they will be removed.

*Note: Please note there are many ways for selecting best variables to build a model, here is presented one more based on visual analysis.*

###### Example 1: Is gender correlated with heart disease? ######

cross_gender=cross_plot(heart_disease, str_input="gender", str_target="has_heart_disease")

![Importance variable analysis with cross\_plot][Importance variable analysis with cross_plot]

Last two plots have the same data source, showing the distribution of `has_heart_disease` in terms of `gender`. The one on the left shows in percentage value, while the one on the right shows in absolute value.

###### How to extract conclusions from the plots? (Short version) ######

`Gender` variable seems to be a **good predictor**, since the likelihood of having heart disease is different given the female/male groups. **it gives an order to the data**.

###### How to extract conclusions from the plots? (Long version) ######

**From 1st plot (%):**

1.  The **likelihood** of having heart disease for males is 55.3%, while for females is: 25.8%.
2.  The heart disease rate for males **doubles** the rate for females (55.3 vs 25.8, respectively).

**From 2nd plot (count):**

1.  There are a total of **97 females**:
    
     *  25 of them have heart disease (25/97=25.8%, which is the ratio of 1st plot).
     *  the remaining 72 have not heart disease (74.2%)
2.  There are a total of **206 males**:
    
     *  114 of them have heart disease (55.3%)
     *  the remaining 92 have not heart disease (44.7%)
3.  Total cases: Summing the values of four bars: 25+72+114+92=**303**.

*Note: What would it happened if instead of having the rates of 25.8% vs. 55.3% (female vs male), they had been more similar like 30.2% vs. 30.6%). In this case variable `gender` it would have been much less relevant, since it doesn’t separate the`has_heart_disease` event.*

###### Example 2: Crossing with numerical variables ######

Numerical variables should be **binned** in order to plot them with an histogram, otherwise the plot is not showing information, as it can be seen here:

###### Equal frequency binning ######

There is a function included in the package (inherited from Hmisc package) : `equal_freq`, which returns the bins/buckets based on the **equal frequency criteria**. Which is *-or tries to-* have the same quantity of rows per bin.

For numerical variables, `cross_plot` has by default the`auto_binning=T`, which automtically calls the `equal_freq`function with `n_bins=10` (or the closest number).

cross_plot(heart_disease, str_input="max_heart_rate", str_target="has_heart_disease")

![Importance variable analysis with cross\_plot][Importance variable analysis with cross_plot 1]

###### Example 3: Manual binning ######

If you don’t want the automatic binning, then set the`auto_binning=F` in `cross_plot` function.

For example, creating `oldpeak_2` based on equal frequency, with 3 buckets.

heart_disease$oldpeak_2=equal_freq(var=heart_disease$oldpeak, n_bins = 3)  
    summary(heart_disease$oldpeak_2)

![Equal frequency binning][]

Plotting the binned variable (`auto_binning = F`):

cross_oldpeak_2=cross_plot(heart_disease, str_input="oldpeak_2", str_target="has_heart_disease", auto_binning = F)

![Importance variable analysis with cross\_plot][Importance variable analysis with cross_plot 2]

###### **Conclusion** ######

This new plot based on `oldpeak_2` shows clearly how: the likelihood of **having heart disease increases** as **oldpeak\_2 increases** as well.*Again, it gives an order to the data.*

###### Example 3: Noise reducing ######

Converting variable `max_heart_rate` into a one of 10 bins:

heart_disease$max_heart_rate_2=equal_freq(var=heart_disease$max_heart_rate, n_bins = 10)  
    cross_plot(heart_disease, str_input="max_heart_rate_2", str_target="has_heart_disease")

![Importance variable analysis with cross\_plot][Importance variable analysis with cross_plot 3]

At a first glance, `max_heart_rate_2` shows a negative and linear relationship, however there are some buckets which add noise to the relationship. For example, the bucket `(141, 146]` has a higher heart disease rate than the previous bucket, and it was expected to have a lower. *This could be noise in data.*

**Key note**: One way to reduce the **noise** (at the cost of **losing** some information), is to split with less bins:

heart_disease$max_heart_rate_3=equal_freq(var=heart_disease$max_heart_rate, n_bins = 5)  
    cross_plot(heart_disease, str_input="max_heart_rate_3", str_target="has_heart_disease")

![Importance variable analysis with cross\_plot][Fancy miniature]

**Conclusion**: As it can be seen, now the relationship is much clean and clear. Bucket *‘N’* has a higher rate than *‘N+1′*, which implies a negative correlation.

**How about saving the cross\_plot result into a folder?**  
Just set the parameter `path_out` with the folder you want -It creates a new one if it doesn’t exists-.

cross_plot(heart_disease, str_input="max_heart_rate_3", str_target="has_heart_disease", path_out="my_plots")

It creates the folder `my_plots` into the working directory.

###### Example 4: `cross_plot` on multiple variables ######

Imagine you want to run cross\_plot for several variables at the same time. To achieve this goal you define a list of strings containing all the variables to use as input in the `cross_plot`, and then, call the function `massive_cross_plot`.

If you want to analyze these 3 variables:

vars_to_analyze=c("age", "oldpeak", "max_heart_rate")

massive_cross_plot(data=heart_disease, str_target="has_heart_disease", str_vars=vars_to_analyze)

**Automatically saving all the results into a folder**  
Same as `cross_plot`, this function has the `path_out` parameter.

massive_cross_plot(data=heart_disease, str_target="has_heart_disease", str_vars=vars_to_analyze, path_out="my_plots")

###### Final notes: ######

*  [Correlation does not imply causation][]
 *  `cross_plot` is good to visualize **linear** relationships, giving it a hint on **non-linear** relationships.
 *  Cleaning the variables help the model to better modelize the data.

#### Part 3: Assessing model performance ####

**Overview**: Once the predictive model is developed with `training`data, it should be compared with `test` data (which wasn’t seen by the model before). Here is presented a wrapper for the **ROC Curve** and **AUC** (area under ROC) and the **KS** (Kolmogorov-Smirnov).

###### Creating the model ######

## Training and test data. Percentage of training cases default value=80%.
    index_sample=get_sample(data=heart_disease, percentage_tr_rows=0.8)
    
    ## Generating the samples
    data_tr=heart_disease[index_sample,]  
    data_ts=heart_disease[-index_sample,]
    
    
    ## Creating the model only with training data
    fit_glm=glm(has_heart_disease ~ age + oldpeak, data=data_tr, family = binomial)

###### ROC, AUC and KS performance metrics ######

## Performance metrics for Training Data
    model_performance(fit=fit_glm, data = data_tr, target_var = "has_heart_disease")

![Model performance, ROC, AUC and KS][Model performance_ ROC_ AUC and KS]

## Performance metrics for Test Data
    model_performance(fit=fit_glm, data = data_ts, target_var = "has_heart_disease")

![Model performance, ROC, AUC and KS][Model performance_ ROC_ AUC and KS 1]

**Key notes**

*  The **higher** the KS and AUC, the **better** the performance is.
    
     *  KS range: from 0 to 1.
     *  AUC range: from 0.5 to 1.
 *  Performance metrics should be **similar** between training and test set.

**Final comments**

*  KS and AUC focus on similar aspects: How well the model distinguishes the class to predict.
 *  ROC and AUC article: [link][]

--------------------

###### End of vignette, some links… ######

[  
![funModeling Github][]Github.  
][_funModeling Github_Github.]Sneak peek into the  `funModeling` “black-box”  *(either for learning or to contribute -code not complex and commented)*:

[R - Data Science Heroes Blog]: http://blog.datascienceheroes.com/package-funmodeling-data-cleaning-variable-importance-analysis-and-model-perfomance/
[R-bloggers]: http://www.r-bloggers.com/
[Fancy miniature]: /images/20220731/dc5a5bf77e854cd48a812d2e72a1ac82.png
[simple-smile.png_zoom_3_w_456]: /images/20220731/e38b7fadb5bf4ff180bc2e16d730bde3.png
[Variable description with df_status]: /images/20220731/34d54cb19e984e14be1b779d76c9cf90.png
[Variable description with df_status 1]: /images/20220731/01943da4bc2140b09e461590719ee975.png
[Variable description with df_status 2]: /images/20220731/b729518de2584888aee5b338dfb97bad.png
[Importance variable analysis with cross_plot]: /images/20220731/a023fd4eba214b64b5ba529b0abd18a2.png
[Importance variable analysis with cross_plot 1]: /images/20220731/883dea048e7047219dcadab4771c6ddd.png
[Equal frequency binning]: /images/20220731/ecdc7fde48af48f99859ea1e4b3f2395.png
[Importance variable analysis with cross_plot 2]: /images/20220731/34a7ade4669c430e9e3350a7cfd3cda6.png
[Importance variable analysis with cross_plot 3]: /images/20220731/0111aad39ca8429eaf208a5bd1c96095.png
[Correlation does not imply causation]: https://en.wikipedia.org/wiki/Correlation_does_not_imply_causation
[Model performance_ ROC_ AUC and KS]: /images/20220731/3836031284934e24bbe33b40f499f6b3.png
[Model performance_ ROC_ AUC and KS 1]: /images/20220731/204156abc5954babb8a9693bc3c292b1.png
[link]: https://en.wikipedia.org/wiki/Receiver_operating_characteristic
[funModeling Github]: /images/20220731/a52be5e8139f47909a5a3d26b558dbd5.png
[_funModeling Github_Github.]: https://github.com/pablo14/funModeling