Fig. 1. Shelf space for instant noodles in a typical Japanese supermarket (a) aisle, and (b) shelf.

Fig. 2. System architecture for a decision system for newly launched products.

Fig. 3. The distribution of survival time of new products. The horizontal axis indicates range of days. The vertical axis indicates counts of brands.

Fig. 4. The two-step transformation procedure for input variables.

Fig. 5. a. The tree obtained from EBONSAI applied on the data set in Table 2. b. Single node produced by ID3 applied to data set in Table 2.

Fig. 6. The l EBONSAI tree for classifying survival status of new instant noodle brands. A higher number implies a higher level of the quantitative variable.

Fig. 7. Comparison of EBONSAI, C4.5, logistic regression, and neural network for predicting survivor and non-survivor brands on performance measures: (a) Recall, (b) Precision, (c) F-measure, and (d) Matthews correlation coefficient.

Fig. 8. Comparing EBONSAI and C4.5 with varying pruning parameter (upper confidence limit of error rate) for Matthews correlation coefficient.

Descriptive statistics of important variables in the model. The week number is represented as (1), (2), and (3)

An example data set to illustrate the shortcoming of learning based on a univariate split

Classification of prediction outcomes. TP: True positives, TN: true negatives, FP: false positive, and FN: false negative