Utilizing Imbalanced Data and Classification Cost Matrix to Predict Movie Preferences

Haifeng Wang

Abstract

In this paper, we propose a movie genre recommendation system based on imbalanced survey data and unequal classification costs for small and medium-sized enterprises (SMEs) who need a data-based and analytical approach to stock favored movies and target marketing to young people. The dataset maintains a detailed personal profile as predictors including demographic, behavioral and preferences information for each user as well as imbalanced genre preferences. These predictors do not include movies' information such as actors or directors. The paper applies Gentle boost, Adaboost and Bagged tree ensembles as well as SVM machine learning algorithms to learn classification from one thousand observations and predict movie genre preferences with adjusted classification costs. The proposed recommendation system also selects important predictors to avoid overfitting and to shorten training time. This paper compares the test error among the above-mentioned algorithms that are used to recommend different movie genres. The prediction power is also indicated in a comparison of precision and recall with other state-of-the-art recommendation systems. The proposed movie genre recommendation system solves problems such as small dataset, imbalanced response, and unequal classification costs.

Introduction

Most modern recommendation engines rely on "Big Data"—millions of ratings and deep metadata about actors, directors, and plot points. However, physical movie retail chains and small-to-medium enterprises (SMEs) often operate with much smaller datasets and need to make localized inventory decisions. Traditional deep learning models struggle with small datasets (approx. 1,000 observations), and standard algorithms often fail when faced with heavily imbalanced data.

Class imbalance across movie genres

As shown in the data, preferences are rarely evenly distributed. For example, while many people favor Science Fiction, very few favor Westerns (a 7:1 imbalance). Standard algorithms often "ignore" the minority class to achieve high overall accuracy, which is ineffective for niche marketing. Furthermore, for a business, misclassifying a potential customer as "uninterested" (a false negative) is more expensive than marketing to someone who isn't interested (a false positive).

Methods

The researchers moved away from movie-centric data and focused entirely on user-centric predictors. The study used a survey dataset of 1,007 observations focused on young people (aged 15–30), including 38 dimensions such as hobbies, personality traits, and spending habits.

Ensemble Classifiers and Cost-Sensitive Learning

The paper compares Bagged Trees, AdaBoost, and GentleBoost. These ensemble methods combine several weak learners to create a robust model resistant to overfitting. The core innovation is the integration of a square cost matrix into the AdaBoost algorithm. By assigning a higher "penalty" to false negatives, the model is forced to prioritize identifying rare but valuable customers.

Confusion matrices for different classification costs

Feature Selection

Using "Out-of-Bag" (OOB) error estimates from bagged trees, the authors identified the most influential predictors. This allowed them to prune the feature set, shortening training time and improving generalization. Interestingly, lifestyle factors like "Online chat" habits and "Outdoors" interest proved to be powerful predictors of movie preference, suggesting that personal habits are strong signals for genre interest.

Results

The proposed AdaBoost model with cost adjustment outperformed standard SVMs and Bagged Trees across several genres. For Science Fiction, it achieved 0.93 Precision and 0.95 Recall. Even for Westerns, despite the heavy 7:1 class imbalance, the model achieved 0.92 Precision and 0.88 Recall.

In experiments with Comedy movies, treating a false negative as five times worse than a false positive significantly improved the model's practical utility. By prioritizing recall (0.86) over precision (0.67), the model ensured that the majority of potential customers were identified, even at the cost of some additional marketing leads. When compared to state-of-the-art recommendation methods like Estimation of Distribution Algorithms (EDA), this SME-focused model remained competitive, proving that smart algorithmic choices can compensate for smaller data volumes.

Conclusion

This research demonstrates that a movie genre recommendation system tailored for SMEs can effectively handle small, imbalanced datasets by utilizing ensemble learning and custom classification cost matrices. While the current model is excellent for localized predictions, future work will focus on expanding the sample size and refining the algorithm to handle broader geographic trends and evolving demographics.

Summary updated

Abstract

In this paper, we propose a movie genre recommendation system based on imbalanced survey data and unequal classification costs for small and medium-sized enterprises (SMEs) who need a data-based and analytical approach to stock favored movies and target marketing to young people. The dataset maintains a detailed personal profile as predictors including demographic, behavioral and preferences information for each user as well as imbalanced genre preferences. These predictors do not include movies' information such as actors or directors. The paper applies Gentle boost, Adaboost and Bagged tree ensembles as well as SVM machine learning algorithms to learn classification from one thousand observations and predict movie genre preferences with adjusted classification costs. The proposed recommendation system also selects important predictors to avoid overfitting and to shorten training time. This paper compares the test error among the above-mentioned algorithms that are used to recommend different movie genres. The prediction power is also indicated in a comparison of precision and recall with other state-of-the-art recommendation systems. The proposed movie genre recommendation system solves problems such as small dataset, imbalanced response, and unequal classification costs.
BibTeX
@article{wang2018utilizing,
  author = {Haifeng Wang},
  title = {{Utilizing Imbalanced Data and Classification Cost Matrix to Predict Movie Preferences}},
  journal = {International Journal of Artificial Intelligence \& Applications},
  year = {2018},
  doi = {10.5121/ijaia.2018.9601},
}
Figure