Dataset preparation

The Joiner is used to merge the outcome variable with the Italian store dataset, which naturally introduced missing values for stores with unknown promotion results. By filtering these rows out with the missing valu enode, the final, clean dataset of only activated stores was fully prepared for the analysis phase

Activated store analysis

Analyzing decision Tree is understandable that distance from a Dentist is the primary splitting criterion, yielding the highest node purity (maximizing gini index) and acting as the key determinant in the data distribution. After that, using the random forest algorithm it is possible to evaluates a multiple sets of trees in order to create a model based on the aggreation of the factors that define the different splits across the trees. By looking at the ROC curve it is understandables that the model is a good predictor respect to a random classifier and also presents a better Area Under the Curve (0.945 AUC) respect to the results of the logistic regression model (0.836 AUC)

Importance of factors

Analyzing the split criteria across the Random Forest allows us to quantify Feature Importance based on cumulative split contribution. The analysis ranks the key drivers of store performance in the following order: Distance to Dentist, Water Hardness IDX, Online Consumption IYA, Nr of Schools, Household avg components, and Average Income per Household. Distance to Dentist, Water Hardness, Online Consumption, and Household avg components exhibit a negative relationship with shop's success (inverse correlation), whereas Nr of Schools and Average Income per Household act as positive performance drivers

Store in which invest in 2026

The provided panel displays the comprehensive classification of all stores. Specifically, the 'Final Outcome' column explicitly delineates the stores recommended for investment from those excluded from the strategy

Logistic regression

Insight: Our logistic regression identified distance to the dentist as the sole significant predictor, revealing that success probability decreases as this distance increases. Although the ROC curve confirms the model is robust and outperforms random chance, we are moving to alternative models because this single-factor explanation lacks sufficient depth

Estimated size price

Comparing the final numbers of the 2026 with 2025