Our original data mainly consists of food composition data and 25 dishes of food.

Food composition data from [4] , containing more than 1200 records, including the sequence number, name, test ingredients, energy, water, protein, fat, vitamins, trace elements, and so on. Figure 1 shows 6 of them.

The 25 dishes of food are from the Sijinuantang restaurant in Wuxi. It contains double fresh with green pepper, original flavor chicken brine, broccoli with mashed garlic, tomato and egg soup and so on. Taking tomato and egg soup as an example, its food composition is shown in Table 1.

Cluster analysis refers to the grouping of physical or abstract objects into multiple classes consisting of similar objects. The goal of cluster analysis is to collect and classify data on similar basis. In different applications, many clustering techniques have been developed, which are used to describe data, to measure the

similarity between different data sources, and to classify data sources into different clusters [5] .

Chen Jian (2007) briefly introduced the concept and principle of cluster analysis [6] . Cluster analysis, also known as group analysis and point group analysis, is an ideal multivariate statistical technique for classification, mainly including partitioning clustering and hierarchical clustering. Partition clustering is also called segmentation clustering. Given a database of M data objects or tuples, the partition clustering is to construct K partition of data, each partition representing a cluster, and k ≤ m. That is, it divides data into K groups and meets the following requirements: 1) each group contains at least one object; 2) each object must belong to and only belong to one group, while second requirements in some fuzzy partitioning techniques can be loosened. The hierarchical clustering divides the database into several levels, then divides and clusters the data of different levels, and outputs a hierarchical classification tree. Hierarchical methods can be divided into cohesive and divisive ones. The method of condensation, also known as a bottom-to-head approach, first takes each object as a cluster, and then merges a similar cluster into a growing cluster until all the objects are in a cluster, or a certain terminative condition is satisfied. The splitting method, also known as the head-to-bottom method, first places all the objects in a cluster and gradually subdivides them into smaller clusters, until each object becomes a cluster, or a certain terminative condition is satisfied.

This thesis uses K-MEANS method [7] , which is one type of partitioning clustering. The idea of the method is: given the number of the classes K, divide N objects into K classes, making the similarity among the objects in the class the most, and the similarity among the classes is the smallest. The calculation of similarity is carried out according to the average value of the objects in a cluster (seen as the center of clustering); that is, each cluster is represented by the average value of the objects in the cluster. The greatest advantage of the method is simplicity and efficiency. The key of the method is the initial center selection and the formula for calculation of the similarity among classes.

We used Python3.6 and sklearn.cluster’s KMeans to cluster analysis, and carried out 20 types and 30 types of cluster analysis respectively. Sklearn is a commonly used Python third party module in machine learning, which can be installed through pip.

The Shannon Wiener diversity index [8] is an index for investigating local diversity in plant communities (alpha diversity), often used in conjunction with the Simpson diversity index. The Shannon Wiener index formula is:

ln was also replaced by log2, Pi for the proportion of individual species in the total number of individuals. In this formula, H = the information content of samples = the diversity index of the community, S = species number, Pi = the proportion of the individuals belonging to the i species in the sample. If the total number of samples of the sample is N, and the number of i species is ni, then Pi = ni/N.

In the Shannon Wiener index, there are two components: species number and distribution uniformity among individuals (equability or evenness). The more evenly distributed individuals are, the greater the H value. If each individual belonged to different species, the diversity index would be the largest; if each individual belonged to the same species, the diversity index would be the smallest.

We use formula (1) to calculate the nutritional balance. In the formula, Pi refers to the total mass of each dish in group i and the percentage of the total mass of all ingredients. A total of 20 groups are generated by clustering.

Take salad ribs as an example. The mass percentages of species number 3, 0, 17, and 19 were 0.9597, 0.0192, 0.0154, 0.0058 respectively. The equilibrium index calculated by formula (1) was 0.3017.

In actual calculation, the quality data (content) of each component of the dish is given by the chef of Wuxi Four Seasons restaurant.

Observing the 20 species, the distribution of 1200 foods in 20 species is shown in the Table 2.

Taking tomato egg soup as an example, on the basis of Table 2, ID (identification number of more than 1200 foods) of the subitems, species number of 20 clusters and species number of 30 clusters were added. For example, in Chart 3 the species number of 20 clusters of eggs is 3.

From Table 3, we can see that the subterm salt and monosodium glutamate of the tomato and egg soup belong to the 17 category of the 20 classes, and the quality is added when the formula (1) is used to calculate the nutritional balance.

Using formula (1) and the method mentioned in the preceding 2.5 sections, the data of nutrient balance of 25 dishes are calculated, as shown in Table 4.