load in necessary packages

overview of the data set

look at the missing values

• validating the dataset for missing values

#>       recipe     calories carbohydrate        sugar      protein     category 
#>            0           52           52           52           52            0 
#>     servings high_traffic 
#>            0          373

data wrangling and exploration

• There are only 2 and 1 recipes of 4 as a snack and 6 as a snack servings, so I’ll rename them to “4” and “6” for simplicity and convert to numerical.
• replace null values of high_traffic with Low
• chicken breast turned to just chicken

inspect the data for the new changes

Data visualisation

• this feature doesn’t have a big influence on target variable because recipes with high traffic are are many for each servings as compared to the those in with low traffic.

Correlations

box plots

• individual plots of both nutrients are shown in the facets below
• looking if there outliers in the nutrients

Histogram

• from the histograms above ,both nutrients are seen to be right skewed

let’s visually inspect single variables

• look at calories

• the data for calories is right skewed as seen from the histogram

• the points in red indicate potential outliers in the data

Conclusion:

the density plots shows that there are no significant depedencies of the traffic and the following numerical features: calories, carbohydrate, protein, sugar, servings.

Modeling data

#> Training cases: 662
#> Test cases: 285
#> # A tibble: 5 × 7
#>   calories carbohydrate sugar protein category     servings high_traffic
#>      <dbl>        <dbl> <dbl>   <dbl> <chr>           <dbl> <fct>       
#> 1   960.           4.4  44.5    12.1  Dessert             1 0           
#> 2   189.           9.54  6.47    0.34 Beverages           6 0           
#> 3   248.          44.7   2.64   19.9  Chicken             1 1           
#> 4     6.23        56.4   5.6     2.12 Lunch/Snacks        6 1           
#> 5    81.0          0.35  1.27    1.19 Beverages           4 0

Train and Evaluate a Binary Classification Model

OK, now we’re ready to train our model by fitting the training features to the training labels (high_trafffic).

Preprocess the data for modelling

• normalize all numerical features
• turn categorical data to numerical data by creating dummy variables

fit the model

#> ══ Workflow ════════════════════════════════════════════════════════════════════
#> Preprocessor: Recipe
#> Model: logistic_reg()
#> 
#> ── Preprocessor ────────────────────────────────────────────────────────────────
#> 2 Recipe Steps
#> 
#> • step_normalize()
#> • step_dummy()
#> 
#> ── Model ───────────────────────────────────────────────────────────────────────
#> Logistic Regression Model Specification (classification)
#> 
#> Computational engine: glm

#> ══ Workflow [trained] ══════════════════════════════════════════════════════════
#> Preprocessor: Recipe
#> Model: logistic_reg()
#> 
#> ── Preprocessor ────────────────────────────────────────────────────────────────
#> 2 Recipe Steps
#> 
#> • step_normalize()
#> • step_dummy()
#> 
#> ── Model ───────────────────────────────────────────────────────────────────────
#> 
#> Call:  stats::glm(formula = ..y ~ ., family = stats::binomial, data = data)
#> 
#> Coefficients:
#>            (Intercept)                calories            carbohydrate  
#>               -3.17385                 0.05534                 0.03697  
#>                  sugar                 protein                servings  
#>               -0.06993                 0.02482                -0.02061  
#>     category_Breakfast        category_Chicken        category_Dessert  
#>                2.42197                 2.95913                 3.81616  
#>  category_Lunch.Snacks           category_Meat  category_One.Dish.Meal  
#>                3.53970                 4.33127                 3.92688  
#>          category_Pork         category_Potato      category_Vegetable  
#>                6.07030                 6.10736                 7.28483  
#> 
#> Degrees of Freedom: 661 Total (i.e. Null);  647 Residual
#> Null Deviance:       880.6 
#> Residual Deviance: 627.8     AIC: 657.8

• all variables whose p value lies below the black line are statistically significant

Let’s take a look at the confusion matrix:

#>           Truth
#> Prediction   0   1
#>          0  94  31
#>          1  26 134

What about our other metrics such as ppv, sensitivity etc?

#> # A tibble: 1 × 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 roc_auc binary         0.844

Model 2 Random forest

Model 2 : evaluation

#>           Truth
#> Prediction   0   1
#>          0  80  22
#>          1  40 143

#> [[1]]
#> # A tibble: 4 × 3
#>   .metric  .estimator .estimate
#>   <chr>    <chr>          <dbl>
#> 1 ppv      binary         0.784
#> 2 recall   binary         0.667
#> 3 accuracy binary         0.782
#> 4 f_meas   binary         0.721
#> 
#> [[2]]
#> # A tibble: 1 × 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 roc_auc binary         0.837
#> 
#> [[3]]

let’s make a Variable Importance Plot to see which predictor variables have the most impact in our model.