Predicting annus mirabilis with machine learning: Turkish movie industry

Abstract

There has recently been an increasing interest in the quantification of success in different fields of human activities. However, most of the research in this field solely focuses on success as a collective phenomenon to be understood in terms of a network structure or overall statistics. Moreover, research in this field rarely attempts to predict success. In this work, we consider real data in the Turkish movie industry, with a focus on individual criteria of success and predict annus mirabilis, also known as the miracle year of the performers, through several machine learning algorithms. We find that this prediction can be achieved best by using a random forest model yielding 92 and 90 percent accuracy for actresses and actors, respectively. Next, we provided a novel q-deformed generalization of the k-Nearest Neighbor (kNN) algorithm, which yields the unnormalized kNN algorithm when the parameter q = 0, and logarithmic normalized kNN algorithm for q = 1. This generalization increases the prediction accuracy of the miracle year by one percent for the optimal value q = −3 of the parameter compared to the random forest algorithm. In addition, we found out that the probability of career length follows an exponential distribution in the intermediate region, hence distinguishing two outlier groups. The former group is formed by those with only one movie credit (almost seventy five percent of the acting guild), and the latter group corresponds to those with a career length exceeding thirty five years (almost one percent). The probability distribution of acting in n movies, on the other hand, is observed to be a power law with an exponent corresponding to Zipf’s law if ordered in rank. This behavior seems to be the norm in the film industry and is a signature of the memory-dependent (namely, popularitydependent) choices of directors and producers. We investigated whether gender plays a role in the Turkish fılm industry, and found out that it does, favoring actors in terms of duration of activity. However, after a certain threshold of career length, namely L = 55 years, it is more likely to find active actresses.