Methods: Survey Design, Recommender Systems, Collaborative Filtering, Content-based Filtering, Deep Learning Approaches, Bayesian Optimization
Abstract
With the growth of the meal kit industry, consumers are faced with a plethora of meal delivery service options. This study surveys current and potential meal kit customers and predicts optimal meal kit choices using statistical and machine learning techniques, including user- and item-based collaborative filtering, content-based filtering, and deep learning approaches. The large number of potential customers who sampled a small percentage of meal kit options presents a challenge in generating optimal predictions. This study implements and discusses the advancement of web-survey methodology and evaluates prediction methods when dealing with sparse data, with the goal of providing an optimal model for predicting consumer meal kit service choices.
Summary
The results of the survey found most respondents have tried very few meal kit services, if any. This led to a sparse dataset for recommendations. Of the different algorithms compared, SVD provided the best overall RMSE and prediction coverage, while performing well when measured against computation times. As the dataset continues to grow, future research will focus on the changes in performance of different algorithms as well as evaluation of non-accuracy based metrics such as serendipity, recommendation diversity, and novelty.
