Högskolan i Skövde

In the digital era and advent of artificial intelligence, digital retailing has emerged as a notable shift in commerce. It empowers e-tailers with data-driven insights and predictive models to navigate a variety of challenges, driving informed decision-making and strategic formulation. While predictive models are fundamental for making data-driven decisions, this thesis spotlights binary classifiers as a central focus. These classifiers reveal the complexities of two real-world problems, marked by their particular properties. Specifically, binary decisions are made based on predictions, relying solely on predicted class labels is insufficient because of the variations in classification accuracy. Furthermore, prediction outcomes have different costs associated with making different mistakes, which impacts the utility.

To confront these challenges, probabilistic predictions, often unexplored or uncalibrated, is a promising alternative to class labels. Therefore, machine learning modelling and calibration techniques are explored, employing benchmark data sets alongside empirical studies grounded in industrial contexts. These studies analyse predictions and their associated probabilities across diverse data segments and settings. The thesis found, as a proof of concept, that specific algorithms inherently possess calibration while others, with calibrated probabilities, demonstrate reliability. In both cases, the thesis concludes that utilising top predictions with the highest probabilities increases the precision level and minimises the false positives. In addition, adopting well-calibrated probabilities is a powerful alternative to mere class labels. Consequently, by transforming probabilities into reliable confidence values through classification with a rejection option, a pathway emerges wherein confident and reliable predictions take centre stage in decision-making. This enables e-tailers to form distinct strategies based on these predictions and optimise their utility.

This thesis highlights the value of calibrated models and probabilistic prediction and emphasises their significance in enhancing decision-making. The findings have practical implications for e-tailers leveraging data-driven decision support. Future research should focus on producing an automated system that prioritises high and well-calibrated probability predictions while discarding others and optimising utilities based on the costs and gains associated with the different prediction outcomes to enhance decision support for e-tailers.

Product returns are not only costly for e-tailers, but the unnecessary transports also impact the environment. Consequently, online retailers have started to formulate policies to reduce the number of returns. Determining when and how to act is, however, a delicate matter, since a too harsh approach may lead to not only the order being cancelled, but also the customer leaving the business. Being able to accurately predict which orders that will lead to a return would be a strong tool, guiding which actions to be taken. This paper addresses the problem of data-driven product return prediction, by conducting a case study using a large real-world data set. The main results are that well-calibrated probabilistic predictors are essential for providing predictions with high precision and reasonable recall. This implies that utilizing calibrated models to predict some instances, while rejecting to predict others can be recommended. In practice, this would make it possible for a decision-maker to only act upon a subset of all predicted returns, where the risk of a return is very high.

Customer churn is one of the most challenging problems for digital retailers. With significantly higher costs for acquiring new customers than retaining existing ones, knowledge about which customers are likely to churn becomes essential. This paper reports a case study where a data-driven approach to churn prediction is used for predicting churners and gaining insights about the problem domain. The real-world data set used contains approximately 200 000 customers, describing each customer using more than 50 features. In the pre-processing, exploration, modeling and analysis, attributes related to recency, frequency, and monetary concepts are identified and utilized. In addition, correlations and feature importance are used to discover and understand churn indicators. One important finding is that the churn rate highly depends on the number of previous purchases. In the segment consisting of customers with only one previous purchase, more than 75% will churn, i.e., not make another purchase in the coming year. For customers with at least four previous purchases, the corresponding churn rate is around 25%. Further analysis shows that churning customers in general, and as expected, make smaller purchases and visit the online store less often. In the experimentation, three modeling techniques are evaluated, and the results show that, in particular, Gradient Boosting models can predict churners with relatively high accuracy while obtaining a good balance between precision and recall. 

Crop yield prediction is an essential one in agriculture. Crop yield protection is the science and practice of handling plant diseases, weeds, and other pests. Accurate information regarding the crop yield history is essential for making decisions regarding agricultural risk management. Many research studies have been undertaken for identifying crop productivity using various data mining techniques. However, the prediction accuracy of crop yields was not improved with minimum time consumption. To overcome the issues, a novel Quantile Regressive Empirical correlative Functioned Deep FeedForward Multilayer Perceptron Classification (QRECF-DFFMPC) Method is proposed for crop yield prediction. QRECF-DFFMPC Method comprises three layers such as input and output layer with one or more hidden layers. The input layer of deep neural learning receives several features and data from the dataset and then sent it to the hidden layer 1. In that layer, Empirical Orthogonal Function is used to select the relevant features with the help of orthogonal basis functions. After that, Quantile regression is used in the hidden layer 2 to analyze the features and produce the regression value for every data point. Then, the regression value of data points is sent to the output layer for improving the prediction accuracy and reducing the time complexity. Experimental evaluation is carried out on factors such as prediction accuracy, precision, and prediction time for several data points and the number of features. The result shows that the proposed technique enhanced the prediction accuracy and precision by 6% and 9% and reduces the prediction time by 32%, as compared to existing works. 

Adding confidence measures to predictive models should increase the trustworthiness, but only if the models are well-calibrated. Historically, some algorithms like logistic regression, but also neural networks, have been considered to produce well-calibrated probability estimates off-the-shelf. Other techniques, like decision trees and Naive Bayes, on the other hand, are infamous for being significantly overconfident in their probabilistic predictions. In this paper, a large experimental study is conducted to investigate how well-calibrated models produced by a number of algorithms in the scikit-learn library are out-of-the-box, but also if either the built-in calibration techniques Platt scaling and isotonic regression, or Venn-Abers, can be used to improve the calibration. The results show that of the seven algorithms evaluated, the only one obtaining well-calibrated models without the external calibration is logistic regression. All other algorithms, i.e., decision trees, adaboost, gradient boosting, kNN, naive Bayes and random forest benefit from using any of the calibration techniques. In particular, decision trees, Naive Bayes and the boosted models are substantially improved using external calibration. From a practitioner’s perspective, the obvious recommendation becomes to incorporate calibration when using probabilistic prediction. Comparing the different calibration techniques, Platt scaling and VennAbers generally outperform isotonic regression, on these rather small datasets. Finally, the unique ability of Venn-Abers to output not only well-calibrated probability estimates, but also the confidence in these estimates is demonstrated.

Association rules are rules that define relationships between items in sales databases. They have been used primarily to organize relevant products in stores in a way to makes them more visible to consumers, which may increase sales and profits. On the other hand, it has been rarely used in recommender systems where algorithms provide instant recommendations by processing consumers' interests that are gathered when browsing online. However, the vast amount of information collected from transaction data saved on backup servers is poorly taken advantage of, because it is not connected to the Internet, although interesting and personalized recommendations can be created after finding the collections of most frequent items, or most interesting rules in such databases. In this paper, we do a critique of the existing research on both recommender systems along with showing their drawbacks, and the association rules with detailed explanations on their advantages. Finally, draw up with several solutions for producing high quality as well as accurate recommendations by applying novel combinations of techniques observed in this research area including the association-rules-based recommender systems.

While consumers value a free and easy return process, the costs to e-tailers associated with returns are substantial and increasing. Consequently, merchants are now tempted to implement stricter policies, but must balance this against the risk of losing valuable customers. With this in mind, data-driven and algorithmic approaches have been introduced to predict if a certain order is likely to result in a return. In this application paper, a novel approach, combining information about the customer and the order, is suggested and evaluated on a real-world data set from a Swedish e-tailer in men's fashion. The results show that while the predictive accuracy is rather low, a system utilizing the suggested approach could still be useful. Specifically, it is reasonable to assume that an e-tailer would only act on predicted returns where the confidence is very high, e.g., the top 1-5%. For such predictions, the obtained precision is 0.918-0.969, with an acceptable detection rate.