Comprehensive Analysis and Optimization of Peach (Prunus persica) Vacuum Drying: Employing Principal Component Analysis, Artificial Neural Networks and the Standard Score Approach

Abstract

Dried peaches are widely consumed as a snack food product and used as an ingredient in cereals as well in chocolate and energy bars. Accordingly, the main objective of this investigation was to optimize the vacuum-drying process for peaches using a combination of three different statistical methods: principal component analysis, the standard score method and an artificial neural network approach. Applied input drying parameters were temperature (50–70 °C), pressure (20–120 mbar) and time (6–10 h), while the investigated output parameters were moisture content, water activity, total color change, phenolic and flavonoid contents and antioxidant activity. It was noted that all investigated output parameters constantly decreased (moisture content, water activity) and increased (total color change, total phenolic and flavonoid contents and antioxidant activity (FRAP, DPPH and ABTS assays)) in accordance with the applied drying temperature. The key variables accounted for 86.33% of data variance based on the PCA results, while the SS and ANN method resulted in the same optimal drying conditions: 60 °C, 70 mbar and 6 h, which indicated the effectiveness of the applied statistical methods.

1. Introduction

The peach (Prunus persica L.) belongs to the Rosaceae family and is an economically important stone fruit species in regions with a temperate climate [1], i.e., the areas with mild to moderate seasonal changes, such as, for example, parts of North America, Europe and Asia. Peach ranks as the third most important fruit grown in temperate zones in the world, after apple and pear [2]. The average annual global production of peaches is 22 million tons [3]. Besides apples and pears, peaches are the third most significant fruit worldwide among deciduous fruit cultivars [4]. In terms of peach production, China occupies the first place in the world, with the production of more than half of the total number of produced peaches worldwide [5]. Except China, Iran, Italy, Spain, the USA and Greece are also significant growers of this seasonal fruit [6]. The chemical composition of peach fruits generally consists of vitamins, minerals, organic acids and also significant quantities of compounds that indicate strong bioactive capacities, such as polyphenols and carotenoids [7]. As secondary metabolites in plants, the presence of polyphenols and carotenoids can be used to estimate the potential for fruit consumption in the daily human diet [8]. Neochlorogenic and chlorogenic acids have been highlighted as predominant phenolic compounds in peaches [9]. However, polyphenols were found to be in a higher concentration in the peach peel compared to their concentration obtained in the edible flesh parts of the peach fruit. This unequal distribution of polyphenols suggests that removing the peel from fruits could cause serious nutrient losses [10]. Carotenoids are natural pigments, which are responsible for most yellow to red shades of fruits and vegetables, including peaches. Besides their properties as food colorants, these compounds are characterized by other functions, of which the most important for human health is their provitamin A activity. Since the human body cannot synthesize vitamin A, carotenoid intake through increased consumption of fruits has gained great importance in recent years [11]. Nutritionally, peaches are recognized for their variety of bioactive compounds, making them a valuable addition to daily consumer diets as part of the new trends with “functional foods” [12].

Most of the produced peaches worldwide are consumed fresh. However, in order to decrease microbiological activity and inevitable deterioration of the overall quality, and thus prolong the relatively short shelf-life of fresh fruits and their commercial availability, peaches can be frozen, dried or processed into products like jams, jellies, juices and baby food. Moreover, peaches can be used as ingredients in the production of a wide range of confectionery and bakery products [13]. Drying is one of the most commonly used food conservation methods, which can be applied in the fruit-processing industry. Achieving the desirable values of moisture content in final dried products provides safety from food spoilage microorganisms. Additionally, reducing water activity makes the food material less susceptible to destructive chemical and enzymatic reactions [14]. The significant reduction in fruit mass and volume during the drying process also facilitates the transport conditions and to a great extent enables the minimization of packaging and storage costs [15]. Various drying techniques have been successfully used for food dehydration, among which the most widely applied is convective drying with the presence of hot air. During air drying, extended exposure of heat-sensitive products to hot air can cause potential degradation of their valuable nutrients, as well as the sensory characteristics [16]. Bound polyphenols present in fruits can be affected and released by applying high temperatures during the drying process. At the same time, these bioactive compounds may be trans-formed into others through oxidation [17]. Improving the quality of dried fruit products by overcoming the previously mentioned disadvantages of conventional drying could be successfully achieved by applying vacuum drying. Due to vacuum application, moisture that initially was present in the fresh fruit could be evaporated at lower temperatures compared to those used with the conventional drying technique [18]. For example, da Silva et al. [19] studied alternative methods to reduce post-harvest losses of nectarine fruit, evaluating the suitability and influence of ultrasound and vacuum, in combination or without, in the drying kinetics and their impact on the quality characteristics of the fruit. Furthermore, combinations of vacuum drying with microwave or infrared drying of fruits were also investigated by Pu and Sun [20] and Zhang et al. [21].

Considering the importance of the potential usage of different dried fruit products as functional or healthy foods, new processing methods have been developed to improve the overall quality of dried products, as numerous studies in the accessible database demonstrated. However, to the best of our knowledge, there is no reported data in the scientific literature that provide a detailed account of how vacuum-drying parameters affect the quality characteristics of peach fruit within a single research study.

Thus, the main objective of the investigation was to optimize the vacuum-drying process of peaches using a combination of the principal component analysis (PCA), the standard score (SS) method and the artificial neural network (ANN) approach. The process parameters considered were drying temperature (ranging from 50 to 70 °C), drying time (6 to 10 h) and pressure (20 to 120 mbar). Various output parameters were analyzed, including moisture content, water activity, total color change (ΔE), total phenolic and flavonoid contents and antioxidant activity (measured using FRAP, DPPH and ABTS assays).

2. Materials and Methods

2.1. Sample

Peach samples were purchased at the local market in Novi Sad (Vojvodina, Serbia) in July 2023. Initially, the stones were removed from each sample, and each peach half was sliced into pieces with a thickness of 3 mm. Subsequently, the samples were frozen and stored at −20 °C before analysis to prevent any potential degradation in the samples.

2.2. Chemicals

The following reagents were purchased from Sigma–Aldrich Chem (Steinheim, Germany): Folin–Ciocalteu reagent, (±)-catechin (purity ≥ 98%), gallic acid (purity ≥ 99%), 2,2-diphenyl-1-picrylhydrazyl (DPPH) (purity ≥ 95%), ABTS (2,2′-azino-bis-(-3-ethylbenzothiazoline-6-sulfonic acid) diammonium salt) (purity ≥ 98%) and TPZT (2,4,6-tris (2-pyridil)-s triazine) (purity ≥ 99%). Trolox (6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid) (purity ≥ 98%) was purchased from Sigma–Aldrich Chem (Steinheim, Germany). All other chemicals and reagents used in the experimental work were of analytical purity.

2.3. Drying Technique

The vacuum-drying procedure was carried out in a vacuum dryer prototype constructed and set up at the Faculty of Technology Novi Sad, University of Novi Sad (Serbia). The previously sliced and frozen samples were placed in the dryer in a frozen state. The samples were not thawed prior to drying to prevent biological changes in the material due to exposure to air. Here, 100 g of sliced peach samples was dried per batch, with the peach slices arranged in a single layer on the shelf. Drying at the central point (as explained in Section 2.4) was conducted until equilibrium moisture was reached, meaning no detectable change in mass after 30 min. A detailed description of the vacuum dryer equipment and vacuum-drying process is presented in the technical solution. Šumić et al. [22,23] also described it in detail in their papers.

2.4. Experimental Design

The response surface methodology (RSM) was applied in this research for the evaluation of the influence of drying parameters (temperature, pressure and time) and their optimization for different responses. An expanded Box–Behnken experimental design with three numeric factors on three levels was applied. The experimental design consisted of 23 experimental runs with five replicates at the central point (60 °C, 70 mbar and 8 h). Independent variables used in the experimental design were temperature (T, 50–70 °C), vacuum pressure (p, 70–120 mbar) and drying time (t, 6–10 h). Furthermore, the experimental design was expanded with six more runs, where different combinations of independent variables were applied in order to provide more detailed information about the influence of vacuum-drying parameters on the investigated physico-chemical properties of dried peaches. The parameters of the applied drying methods are shown in

Table 1. Vacuum drying conditions set as Box–Behnken experimental design with 17 + 6 runs.

Sample	T (°C)	Code	p (mbar)	Code	t (h)	Code
1	60	0	120	1	6	−1
2	50	−1	70	0	10	1
3	70	1	70	0	10	1
4	60	0	120	1	10	1
5	60	0	70	0	8	0
6	60	0	20	−1	6	−1
7	50	−1	120	1	8	0
8	60	0	70	0	8	0
9	50	−1	20	−1	8	0
10	50	−1	70	0	6	−1
11	60	0	70	0	8	0
12	70	1	20	−1	8	0
13	70	1	120	1	8	0
14	70	1	70	0	6	−1
15	60	0	20	−1	10	1
16	60	0	70	0	8	0
17	60	0	70	0	8	0
18	70	1	70	0	8	0
19	50	−1	70	0	8	0
20	60	0	20	−1	8	0
21	60	0	120	1	8	0
22	60	0	70	0	6	−1
23	60	0	70	0	10	1

.

2.5. Analysis

2.5.1. Extract Preparation

The extracts for the subsequent analysis of phenolic compounds were prepared following the method of González-Gómez et al. [24] with some modifications. Dried peach samples were initially ground in an IKA A11 basic blender, and then approximately 5.0 g of each peach sample was combined with 50 mL of the extraction solvent (95% methanol). After the extraction (24 h), extracts were first filtered and then stored in the refrigerator before analysis. These extracts were subsequently used to determine the total phenolic content [25], total flavonoid content [26] and antioxidant activity (FRAP, DPPH and ABTS assays).

Antioxidant Activity—FRAP Assay

The slightly modified method presented by Benzie and Strain [27] was used to measure samples’ ability to reduce Fe³⁺ in the FRAP assay. FRAP reagent was prepared freshly from 300 mM acetate buffer (pH = 3.6), 10 mM 2,4,6-tris(2-pyridyl)-s-triazine (TPZT) in 40 mM HCl solution and 20 mM aqueous FeCl₃ solution. Then, the prepared solutions were mixed in a ratio of 10:1:1 (v/v/v). Furthermore, rediluted extracts and FRAP reagent were mixed (0.1 mL + 1.9 mL) and stored to incubate in the dark at 37 °C for 10 min. Measurements were then performed at 593 nm with a UV-VIS spectrophotometer (LLG-uniSPEC 2, UV/VIS-Spectrometer, producer LLG Labware, Meckenheim Germany). A standard curve had been previously established (concentration range 7.81–500 mg/L). The results were finally expressed as mg Fe²⁺/g dry weight.

Antioxidant Activity—DPPH Assay

A modified method of that originally presented by Brand-Williams et al. [28] was used to measure the samples’ ability to scavenge 2,2-diphenyl-1 picrylhydrazyl free radicals (DPPH). A methanol solution of DPPH reagent (65 µM) was prepared freshly and adjusted with methanol to reach an absorbance of 0.70 (±0.02). Previously diluted extracts and DPPH reagent were then mixed (2.9 mL + 0.1 mL) in plastic cuvettes (10 mm) and incubated at room temperature (for 60 min). Measurement was carried out at a wavelength of 517 nm using a UV-VIS spectrophotometer (Cary 60, Agilent Technologies, Santa Clara, CA, USA). A standard curve had been previously established (concentration range 6.25–200 mg/L). The results obtained were expressed as mg of Trolox equivalents per gram of dry weight.

2.5.2. Antioxidant Activity—ABTS Assay

Finally, the modified method originally described by Re et al. [29] was used to measure the ABTS free radical scavenging ability of samples. The ABTS stock solution was freshly prepared by combining equal parts (1:1, v/v) of a 2.45 mM potassium persulfate aqueous solution and a 7 mM ABTS (2,2′-azino-bis-(-3-ethylbenzothiazoline-6-sulfonic acid) diammonium salt) aqueous solution, then kept in darkness at ambient temperature for 16 h. The stock solution was subsequently diluted with a 300 mM acetate buffer (pH = 3.6) to reach an absorbance of 0.70 (±0.02). Previously diluted extracts and ABTS reagent were combined (0.1 mL + 2.9 mL) and left in darkness at room temperature for 300 min. Measurements were conducted at 734 nm using a UV–VIS spectrophotometer (Cary 60, Agilent Technologies). A standard curve had been previously established (concentration range 6.25–200 mg/L). Results were ultimately expressed as mg Trolox equivalents per gram of dry weight.

2.5.3. Total Phenol Content (TPC)

The TPC in all samples was determined spectrophotometrically (LLG- by the Folin–Ciocalteau method [25], using gallic acid as a standard). The content of TPC was expressed as the gallic acid equivalent (mg GAE/100 g DW).

2.5.4. Total Flavonoid Content (TFC)

The TFC in all samples was determined spectrophotometrically, using the colorimetric method with aluminum chloride [26]. The content of TF was expressed as the catechin equivalent (mg CE/100 g DW).

2.5.5. Moisture Content (MC)

The moisture content was assessed by drying the samples at 105 °C until a constant weight was achieved. Results were expressed in percentages.

2.5.6. Water Activity (a_w)

Water activity was measured by placing approximately 2.5 g of finely chopped and dried peach samples into the sample chamber of a LabSwift a_w-meter (Novasina, Lachen, Switzerland), set to 25 °C. The a_w values were recorded after reaching equilibrium.

2.5.7. Total Color Change (ΔE)

The CIE L*a*b* color coordinates were determined using a MINOLTA Chroma Meter CR-400 (Minolta Co., Ltd., Osaka, Japan). The surface color of the samples was assessed based on L (lightness level), a (red–green spectrum) and b (yellow–blue spectrum) values. Samples were positioned on the measuring head of the Chroma Meter, and color readings were taken for each prepared sample. Calibration was carried out using a standard white reference. Total color change was analyzed relative to the fresh peach sample, specifically showing the difference between the fresh and dried peach samples for each drying method individually.

2.6. Statistical Analysis

2.6.1. Principal Component Analysis (PCA)

Principal component analysis (PCA) was conducted to interpret and organize the results, i.e., to highlight the differences and groupings among the peach samples dried in different drying conditions. In this method, initial variables were transformed by using eigenvalue decomposition to a set of linearly uncorrelated variables called principal components (PCs). The PCA assessment was conducted using Statistica 10.0 software (StatSoft Inc., Tulsa, OK, USA).

2.6.2. Standard Scores (SSs)

The ranking of the samples was performed by comparing the raw data of each assay with the extreme values, following the methodology described by Brlek et al. [30]. The ranking process considered two criteria: “the higher, the better” for parameters such as total phenolic content and total flavonoid content, as well as antioxidant activity (FRAP, DPPH and ABTS assays) according to Equation (1), and “the lower, the better” for moisture content, a_w value and total color change according to Equation (2).

$$\overline{x_i}={\displaystyle \frac{x_i-x_{\min }}{x_{\max }-x_{\min }}},\hspace{1em}\forall i\text{ in the case of “the higher, the better” criteria}$$

(1)

$$\overline{x_i}=1-{\displaystyle \frac{x_i-x_{\min }}{x_{\max }-x_{\min }}},\hspace{1em}\forall i\text{ in the case of “the lower, the better” criteria}$$

(2)

where $\overline{x_i}$ represents the normal score for parameter x, and x_i represents the raw data.

The equations used in this ranking approach involved comparing the raw data (represented by x_i) with the respective extreme values. The SS value is obtained by averaging the normalized scores of a sample across various measurements. It represents a unique value derived from a combination of data obtained from different measuring methods. This approach has the advantage of facilitating future comparisons by allowing the inclusion of additional sets of vacuum-dried peach samples in the analysis.

2.6.3. Artificial Neural Network (ANN)

In this study, a three-layer multi-layer perceptron model (MLP) was employed for modeling, as it was proven effective in approximating nonlinear functions in previous research [31]. The architecture of the artificial neural network (ANN) utilized a feedforward design and implemented the backpropagation training algorithm.

To train the ANN model, the experimental dataset was randomly divided into two groups: training and testing data, representing 70% and 30% of the experimental data, respectively. During the training phase, the network was repeatedly exposed to the input data [32], as defined by Equation (3):

$$Y=f_1(W_2·f_2(W_1·X+B_1)+B_2)$$

(3)

In order to optimize the ANN model, the Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm was utilized, which is an iterative method for solving unconstrained nonlinear optimization problems. The training process involved a total of 100,000 epochs, representing the number of training steps. The weights and biases associated with the hidden and output layers were organized in matrices W₁, B₁, W₂ and B₂, respectively [33]. These weights were defined during the ANN learning process, where optimization procedures were implemented to minimize the error between the network’s output and the experimental values [34].

To assess the performance of the constructed nonlinear models, standard statistical tests were conducted, including the coefficient of determination (r²), reduced chi-square (χ²), mean bias error (MBE), root mean square error (RMSE), mean percentage error (MPE), sum of squared errors (SSE) and average absolute relative deviation (AARD). These commonly used parameters can be calculated as follows (Equations (4)–(9)) [35]:

$${\chi }^2={\displaystyle \frac{{\displaystyle \sum _{i=1}^N{(x_{\exp ,i}-x_{pre,i})}^2}}{N-n}}$$

(4)

$$RMSE={\left[{\displaystyle \frac{1}{N}}·{\displaystyle \sum _{i=1}^N{(x_{pre,i}-x_{\exp ,i})}^2}\right]}^{1/2}$$

(5)

$$MBE={\displaystyle \frac{1}{N}}·{\displaystyle \sum _{i=1}^N(x_{pre,i}-x_{\exp ,i})}$$

(6)

$$MPE={\displaystyle \frac{100}{N}}·{\displaystyle \sum _{i=1}^N(\frac{\left|x_{pre,i}-x_{\exp ,i}\right|}{x_{\exp ,i}})}$$

(7)

$$SSE={\displaystyle \sum _{i=1}^N{(x_{pre,i}-x_{\exp ,i})}^2}$$

(8)

$$AARD={\displaystyle \frac{1}{N}}·{\displaystyle \sum _{i=1}^N\left|\frac{x_{\exp ,i}-x_{pre,i}}{x_{\exp ,i}}\right|}$$

(9)

where x_exp,i are experimental values and x_pre,i are the model predicted values, and N and n are the numbers of observations and constants accordingly.

2.6.4. Global Sensitivity Analysis—Yoon’s Interpretation Method

Additionally, the relative importance of the input variables (temperature, pressure and time) on the output variables (moisture content, water activity, total color change, phenolic and flavonoid content and antioxidant activity) was determined using the method described by Yoon et al. [36]. In this approach, the relative importance (RI_ij) of the i-th input variable on the j-th output is calculated based on the weights (w_ik) between the i-th input and the k-th hidden neuron, as well as the weights (w_kj) between the k-th hidden neuron and the j-th output.

Equation (10) was applied to evaluate the direct effect of the input parameters on the output variables, based on the weighting coefficients within the ANN models:

$$R{I}_{ij}(\%)={\displaystyle \frac{{\displaystyle \sum _{k=0}^n(w_{ik}·w_{kj})}}{{\displaystyle \sum _{i=0}^m\left|{\displaystyle \sum _{k=0}^n(w_{ik}·w_{kj})}\right|}}}·100\%$$

(10)

where w—presents the weights of the ANN models, i—input variable, j—output variable, k—hidden neuron, n—number of hidden neurons, m—number of inputs.

3. Results and Discussion

Experimentally obtained values of all investigated output parameters (moisture content (MC), water activity (a_w), total color change (ΔE), total phenolic content (TPC), total flavonoid content (TFC) and antioxidant activity (FRAP, DPPH and ABTS assays)) are presented in

Table 2. Experimentally obtained values of MC (%), a_w, ΔE, TPC (mg GAE/100 g DW), TFC (mg CE/100 g DW) and antioxidant activity (FRAP (mg Fe²⁺/g DW), DPPH (mg Trolox/g DW) and ABTS (mg Trolox/g DW) assays).

Sample	MC	aw	ΔE	TPC	TFC	FRAP	DPPH	ABTS
1	11.89	0.408	36.12	393.46	41.42	3.64	2.36	6.97
2	11.38	0.420	35.93	573.85	57.35	3.89	3.92	8.91
3	8.04	0.353	42.64	736.19	93.26	4.90	5.94	9.17
4	10.34	0.424	35.71	520.75	65.18	4.04	4.06	8.24
5	7.00	0.267	38.02	494.00	48.39	3.52	1.08	8.06
6	8.86	0.284	42.52	599.57	73.95	4.98	5.46	12.67
7	15.58	0.492	35.85	371.09	44.82	2.72	1.33	4.22
8	10.12	0.348	35.58	529.89	64.02	3.68	4.10	7.90
9	13.87	0.433	44.98	688.23	79.29	4.54	6.48	14.36
10	24.95	0.702	34.73	489.95	49.82	3.35	2.72	6.18
11	11.61	0.373	41.01	430.30	48.89	3.22	1.98	5.18
12	9.67	0.326	36.51	597.70	88.99	3.54	4.82	12.66
13	9.74	0.335	38.84	448.86	59.29	3.50	1.48	7.93
14	16.63	0.500	39.09	682.95	86.75	4.96	5.34	13.62
15	15.97	0.471	40.98	563.43	62.62	4.41	6.24	7.30
16	10.89	0.370	42.10	709.32	93.73	5.32	4.36	14.93
17	15.00	0.475	42.59	559.21	53.47	4.58	4.84	9.49
18	11.43	0.348	45.82	731.61	102.39	4.97	5.56	14.54
19	22.10	0.598	34.32	348.29	37.65	2.37	1.22	2.53
20	12.16	0.361	40.75	741.95	94.57	5.85	7.18	14.83
21	12.29	0.455	34.67	601.68	69.18	4.62	5.83	9.32
22	15.74	0.467	40.99	791.25	90.09	7.01	8.26	15.29
23	11.43	0.350	35.50	503.39	52.84	3.95	2.26	7.92
F test	185.78	86.98	14.03	102.53	142.87	78.70	503.96	264.61
p-value	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
Polarity	−	−	−	+	+	+	+	+
Weight	0.1	0.1	0.3	0.1	0.1	0.1	0.1	0.1

Polarity: ‘+’ = the higher, the better criteria, and ‘−’ = the lower, the better criteria. Weight: the technological importance of the variable.

. The influence of each input drying parameter (temperature, pressure and time) is presented in Figure 1a–c.

The ANOVA calculation provided presents the F values and p-values for parameters, including MC, a_w, ΔE, TPC, TFC, FRAP, DPPH and ABTS. F values: F values were high for all parameters, indicating that there was likely a statistically significant difference between groups in the experimental results for each parameter. All p-values were less than 0.001. This strongly suggested that the observed differences across groups were statistically significant for each parameter (statistically significant at p < 0.001 level).

The results analyzed using the expanded Box–Behnken experimental design did not give satisfactory outcomes, leading to the decision to interpret each input parameter (temperature, pressure and time) individually and in detail. This approach provided a more detailed understanding of how each factor influenced the outcome. Additionally, alternative statistical methods such as ANN, SS and PCA were applied to improve data analysis and draw more accurate conclusions. These methods helped to better model the relationships between variables and to improve the overall interpretability of the results.

It could be seen (Figure 1a) that all investigated output parameters were constantly increasing, i.e., decreasing in accordance with the applied drying temperature. Precisely, MC and a_w values in dried peach samples decreased as the drying temperature increased, so the minimum values of these two parameters were noted at the highest applied temperature, 70 °C. Accordingly, ΔE, TPC, TFC and all antioxidant activity parameters (FRAP, DPPH and ABTS assays) increased as the drying temperature increased, and all maximum values of these parameters were obtained at the maximum investigated drying temperature, 70 °C. This could be explained by the fact that food processing might accelerate the release of a greater number of bound phenolic compounds due to the breakdown of cellular components. Although the disruption of cell walls may also trigger the release of oxidative and hydrolytic enzymes that could destroy the antioxidant activity in fruits, the high temperature of the hot-air drying process would deactivate these enzymes and prevent the loss of phenolic compounds, thereby leading to an increase in antioxidant activity [37]. Regarding the color degradation, it is accelerated with an increase in temperature. These reactions are caused by the rise in material temperature during drying and the presence of oxygen, which occurred here as the absolute vacuum could not be achieved. The increase in total color change ΔE during the drying process is caused by various factors, such as chemical reactions, for example, the Maillard reaction [38], as well as the degradation of pigments contained in the fruit [39]. The application of higher drying temperatures affects both the increased intensity of chemical reactions and the degradation of pigments, which directly contributes to the increased ΔE values. As presented in the paper by García-Moreira et al. [40], the peach slices dried by the solar method at 3 m/s presented the highest redness color during the total drying time, probably due to the lower drying temperature (37.2 °C). In the same paper [40], it was shown that the percentage loss in TPC content as affected by the drying method ranged from 54.16 to 72.5% compared to the fresh product; as well as that, the TPC content of forced convection at different temperatures showed different behaviors, and this method at 45 °C presented the highest value of TPC.

When analyzing the influence of the pressure on each output parameter during the drying process (Figure 1b), it was noted that MC, a_w and ΔE were the highest at the pressure of 70 mbar, while the first two parameters (MC and a_w) were lowest at the lowest applied pressure, 20 mbar. Furthermore, the minimum applied pressure, 20 mbar, resulted in the samples with the highest values of TPC, TFC and all investigated antioxidant activity parameters. These results are in accordance with the main principles of vacuum drying, where a vacuum is applied to accelerate the removal of water and reduce the drying temperature, thereby protecting heat-sensitive food components. Thus, vacuum drying successfully dries sensitive fruit with minimal changes in volume, chemical composition and sensory properties [41,42]. Accordingly, different values of the applied pressure differently influenced the output parameters investigated in this research; increasing the pressure from 20 to 120 mbar (decreasing the vacuum) has the influence of decreasing the contents of bioactive compounds and antioxidant activity, and on the other hand, of in increasing the moisture content and water activity, which was thoroughly examined and confirmed in previous published papers [43,44,45]. The influence of the applied vacuum during fruit drying was also investigated by other authors, such as [46], where effects of different drying methods and temperatures on the drying behavior and quality attributes of cherry laurel fruit were investigated, and based on the results, it was proposed that ultrasound-assisted vacuum drying could offer an alternative method to hot-air drying due to the higher bioactive compound retention and rehydration ratio, shorter drying time, less color change and shrinkage formation. Similarly, in the research [47], where the authors studied the effects of vacuum drying, hot-air drying and freeze drying on polyphenols and the antioxidant capacity of lemon pomace aqueous extracts, the influence of the vacuum application was presented through the obtained results, and it was concluded that vacuum drying at 90 and 70 °C for 7 and 18 h, respectively, is a good method for the preservation of the total flavonoid content.

Finally, as was expected, the lowest values of MC and a_w in dried peach samples were obtained in samples dried the longest, i.e., 10 h (Figure 1c). At the same drying time, the lowest ΔE was noted, while the highest TPC, TFC and antioxidant activity values were observed in samples dried for 6 h, i.e., at the shortest drying time. The influence of the drying time in the case of vacuum drying of fruit had not been previously investigated, so an approximate explanation is presented instead based on comparison with the results presented in the paper by El-Beltagi et al. [48], where it was noted that the maximum (13 h) drying time was recorded in peaches dried by the open sun method, followed by oven drying (10 h), while the shortest drying time (7 h) was observed in peaches dried by the solar-drying method with a 0.5 cm slice thickness. Also, in the same paper [48], it was noted that the minimum moisture content reduction (8.45%) was recorded in open sun drying with a 13 h drying duration.

3.1. Principal Component Analysis (PCA)

Principal component analysis was employed to explore the relationships among different samples. It has already been widely used in the statistical analysis of fruit drying, such as, for example, in the paper by Francini et al. [49], where apple fruit was dried and where PCA analysis produced two significant components that, together, explained 86.33% of the total variance in the dataset. The results of the PCA in this research are visualized in Figure 2, where the proximity of points on the PCA graph indicates similarity in patterns [50]. The direction of the vectors in the factor space shows the trends of the observed variables, while the length of the vectors indicates the strength of the correlation between the fitting value and the variable [50].

The first two principal components (PCs) accounted for 86.33% of the total variance in the recorded data, specifically 63.55% for the first PC and 22.78% for the second PC, for variables such as moisture content, water activity and total color change.

The projection of variables on the factor plane indicated that ΔE contributed negatively to the first principal component (PC 1) by 10.59% based on correlation, along with TF (16.94%), ABTS (17.24%), TP (18.17%), FRAP (16.13%) and DPPH (14.70%%). For the second principal component (PC 2), MC and a_w made negative contributions of 44.79% and 44.37%, respectively (Figure 2).

3.2. Standard Scores (SSs)

The objective of identifying optimal values for the output variables, based on the processing variables (drying temperature, time and pressure), was achieved through the calculation of standard scores. The standard scores for the parameters are presented in Table 2 and provide a comprehensive overview of these variables.

The standard score (SS) was computed by summing the normalized scores of each variable, which were then multiplied by their respective weights. By maximizing the SS function (T, t, p), both the optimal processing parameters and the corresponding optimal values for the output variables can be determined. A higher SS value indicated a stronger likelihood that the tested processing parameters were optimal, with a value of 1 indicating a high level of optimality.

The highest SS value (0.785) was observed in sample 22 (60 °C, 70 mbar and 6 h) (Figure 3), which aligned with the results obtained with ANN modeling, where this sample was indicated as optimal for peach fruit drying.

3.3. Artificial Neural Network (ANN)

In this study, several artificial neural network (ANN) models were developed to predict the values of output variables (moisture content, water activity, total color change, total phenolic and flavonoid contents and antioxidant activity (measured using FRAP, DPPH and ABTS assays)) based on input variables such as T, t and p. The performance and structure of the ANN were significantly influenced by the initial assumptions made for matrix parameters such as biases and weight coefficients. These parameters were crucial in constructing and fitting the ANN to experimental data. Additionally, the number of neurons in the hidden layer had an impact on the model’s performance. To address this issue, 100,000 topologies were tested to eliminate any random correlations caused by initial assumptions and random weight initialization. This rigorous approach resulted in achieving the highest r² value during the training phase when using eight hidden neurons for moisture content prediction (Figure 4a).

For the ANN model, the training cycle was conducted for 100 epochs, and the corresponding results, including training accuracy and error (loss), are presented in Figure 4b. The training accuracy exhibited a consistent increase with each training cycle until it reached a relatively stable value around the 40–50th epochs. Using more than 50 epochs for training could potentially lead to significant overfitting. Therefore, to ensure high model accuracy without the risk of overfitting, it would be suitable to use 50 as a sufficient number of epochs (Figure 4b).

The performance levels of the ANN models were assessed, and it was found that the optimal number of neurons in the hidden layers for output variables’ prediction was 6–10, as shown in

Table 3. Artificial neural network model summary (performance and errors), for training, testing and validation cycles.

	Net. Name	Training Perf.	Test Perf.	Train. Error	Test Error	Train. Algorithm	Error Funct.	Hidden Activation	Output Activation
MC	MLP 3-8-1	0.917	0.954	0.720	1.566	BFGS 125	SOS	Tanh	Identity
aw	MLP 3-10-1	0.899	0.940	0.000	0.004	BFGS 80	SOS	Tanh	Identity
TCC	MLP 3-10-1	0.870	0.969	0.772	2.002	BFGS 204	SOS	Exponential	Identity
TP	MLP 3-10-1	0.879	0.837	938.869	2988.411	BFGS 119	SOS	Tanh	Identity
TF	MLP 3-7-1	0.832	0.596	31.308	112.616	BFGS 102	SOS	Logistic	Identity
FRAP	MLP 3-9-1	0.875	0.807	0.065	0.111	BFGS 180	SOS	Tanh	Identity
DPPH	MLP 3-9-1	0.885	0.990	0.236	0.672	BFGS 74	SOS	Tanh	Identity
ABTS	MLP 3-8-1	0.830	0.576	1.132	0.759	BFGS 168	SOS	Exponential	Identity

, resulting in high values of the coefficient of determination (r²) during the training cycle of ANN building, as shown in

Table 4. Calculated parameters for the applied artificial neural networks.

ANN	χ2	RMSE	MBE	MPE	SSE	AARD	r2
MC	1.506	1.200	0.000	4.235	33.130	4.235	0.917
aw	0.001	0.031	0.000	2.961	0.022	2.961	0.899
TCC	1.614	1.242	0.000	1.365	35.497	1.365	0.870
TP	2.0 × 103	43.333	0.000	2.844	4.3 × 104	2.844	0.879
TF	65.463	7.913	0.000	4.648	1.4 × 103	4.648	0.832
FRAP	0.136	0.361	0.000	3.773	2.991	3.773	0.875
DPPH	0.493	0.687	0.000	15.091	10.846	15.091	0.885
ABTS	2.366	1.504	0.000	6.400	52.054	6.400	0.830

.

The predicted values from the ANN models exhibited good agreement with the desired values, as reflected by the r² values. The SOSs obtained with the ANN models were within the same order of magnitude as the experimental errors reported in the literature. The training error for prediction of TPC, TFC, TMA and IC₅₀, using the ANN model reached 0.072, the testing error was 0.206 and the validation error was 0.514, while the RMSE values were 6.015, 6.483, 1.247 and 0.199 [51]. The r² values for prediction of TPC, TFC, TMA and IC₅₀ reached 0.981, 0.870, 0.691 and 0.901 [51]. The RMSE values for ANN prediction of the dielectric constant in vegetable oils were 0.2–0.3 [52], while the RMSE values for ANN prediction of the speed of sound in vegetable oils reached values between 1.0 and 1.4 [52].

The r² values between the experimental and ANN model outputs for various parameters, including moisture content, water activity, total color change, total phenolic and flavonoid contents and antioxidant activity (measured using FRAP, DPPH and ABTS assays), were 0.917, 0.899, 0.870, 0.879, 0.832, 0.875, 0.885 and 0.830, respectively, during the training period.

To evaluate the quality of the model fit, several metrics were considered, and a good fit to the experimental values was indicated by a high r² value close to 1 and lower values for the other tests (χ², RMSE, MBE, MPE, SSE and AARD) [53], as shown in Table 4.

Based on the computed parameters for the applied ANN (Table 4), it could be observed that the coefficient of determination (r²) for all output parameters varied between 0.830 (ABTS) and 0.917 (MC), suggesting that the ANN could effectively describe the behavior of peach during the vacuum-drying process.

3.3.1. Optimal Sample

Input parameters: T = 60 °C, p = 70 mbar, t = 6 h.

Output parameters: moisture content = 15.74%; water activity = 0.467; total color change = 40.99; TP = 791.25 mg GAE/100 g DW; TF = 90.09 mg CE/100 g DW; FRAP = 7.01 mg Fe²⁺/g DW; DPPH = 8.26 mg Trolox/g DW; ABTS = 15.29 mg Trolox/g DW.

3.3.2. Global Sensitivity Analysis—Yoon’s Interpretation Method

These results were consistent with the evaluation of standard score analysis. The method of weight partitioning by van Griensven et al. [54] allowed for determining the relative importance levels of input parameters for output parameters of the ANN. These weights between the input, hidden and output layers were used to calculate the relative importance of each input parameter separately. The relative influence levels of input parameters on moisture content, water activity, total color change, total phenolic and flavonoid contents as well as antioxidant activity (FRAP, DPPH and ABTS assays) are presented in Figure 5.

In Figure 5, it can be seen that in the case of MC, TPC and all three antioxidant assays, the temperature had a negative relative influence, while in the case of aw, ΔE and TFC, this influence was positive. In case of the pressure, it was noted that there was no relative influence of pressure in case of MC, ΔE and TFC; a positive relative influence of pressure was noted in case of a_w and TPC, while for the rest of the parameters (all three antioxidant assays), the relative influence of pressure was positive. Finally, the relative influence of the time as an input parameter was recognized as positive in almost all investigated output parameters except for the DPPH assay, where it was negative, and TFC, where no relative influence of pressure was noted.

Generally, based on these results, it can be concluded that temperature has the greatest influence on most of the parameters, while time and pressure have a lesser impact, but that the pressure is often negatively associated with antioxidant parameters.

4. Conclusions

Based on the results obtained from the detail analysis of the influence of each input parameter individually, it was observed that all investigated output parameters were constantly decreasing (moisture content, water activity), i.e., increasing (total color change, total phenolic and flavonoid contents and antioxidant activity (FRAP, DPPH and ABTS assays)) in accordance with the applied drying temperature. The influence of pressure in the vacuum-drying chamber during drying also had a significant influence, with the minimum applied pressure of 20 mbar resulting in samples with the highest values of total phenolic and flavonoid contents and all investigated antioxidant activity parameters. When it came to the influence of drying time, it was noted that the lowest values of moisture content and water activity in dried peach samples were obtained in samples dried the longest, i.e., 10 h.

According to the results of PCA, it was concluded that the first two principal components (PCs) accounted for 86.33% of the variance. Based on the results of the applied standard score method, it was observed that the highest SS value was 0.785, noted in peach sample 22 dried at 60 °C, 70 mbar and 6 h, and furthermore this result was in accordance with the results obtained with ANN modeling, where the optimal sample was a peach sample dried in the same conditions and where the output parameters were moisture content = 15.74%; water activity = 0.467; total color change = 40.99; TP = 791.25 mg GAE/100 g DW; TF = 90.09 mg CE/100 g DW; FRAP = 7.01 mg Fe²⁺/g DW; DPPH = 8.26 mg Trolox/g DW; ABTS = 15.29 mg Trolox/g DW.

Based on all the results obtained in this investigation and the defined input parameters used to achieve the optimal dried peach sample, it is evident that the applied statistical methodology demonstrates effectiveness. These results provide a good basis for improving the quality of the peach drying.