CLASSIFICATION OF COMMERCIAL OMANI HALWA BY PHYSICO-CHEMICAL PROPERTIES AND INSTRUMENTAL TEXTURE PROFILE ANALYSIS (TPA)

Abstract

Commercially available Omani halwa from the local market was classified based on its composition and instrumental texture profile. The products were grouped into four classes based on cluster analysis. None of the attributes of the instrumental Texture Profile Analysis (TPA) were correlated significantly with the ash, colour and pH values (p<0.05), while specific attributes did correlate with specific aspects of the chemical composition. All TPA attributes were significantly correlated with specific types of fatty acid. Hardness was correlated significantly with the moisture and sugar contents, while adhesiveness was correlated significantly with the moisture, sugar and non-sugar carbohydrate contents (p<0.05). Springiness was correlated with sugar and non-sugar carbohydrate. The firmness and chewiness were correlated with moisture and the total fat, saturated and unsaturated fat contents (p<0.05). Cohesiveness-1 was correlated only with the protein, total fat, saturated and un-saturated fat contents, while cohesiveness-2 was correlated with sugar, total fat, saturated and unsaturated fat (p<0.05). The resilience was correlated with the total fat, saturated and unsaturated fatty acids, while gumminess was only correlated with the moisture content. The link between TPA and physico-chemical characteristics was established using multivariate matrix correlations and revealed that the pattern of texture attributes was mostly linked to the moisture and fat contents. Principal component analysis was carried out to identify the main physico-chemical properties and TPA attributes of the four classes of halwa as determined by cluster analysis. The four classes of halwa could be characterised as soft-resilient, soft-springy-cohesive, soft-springy and hard-chewy. [PUBLICATION ABSTRACT]

Headnote

ABSTRACT

Commercially available Omani halwa from the local market was classified based on its composition and instrumental texture profile. The products were grouped into four classes based on cluster analysis. None of the attributes of the instrumental Texture Profile Analysis (TPA) were correlated significantly with the ash, colour and pH values (p<0.05), while specific attributes did correlate with specific aspects of the chemical composition. All TPA attributes were significantly correlated with specific types of fatty acid. Hardness was correlated significantly with the moisture and sugar contents, while adhesiveness was correlated significantly with the moisture, sugar and non-sugar carbohydrate contents (p<0.05). Springiness was correlated with sugar and non-sugar carbohydrate. The firmness and chewiness were correlated with moisture and the total fat, saturated and unsaturated fat contents (p<0.05). Cohesiveness-1 was correlated only with the protein, total fat, saturated and un-saturated fat contents, while cohesiveness-2 was correlated with sugar, total fat, saturated and unsaturated fat (p<0.05). The resilience was correlated with the total fat, saturated and unsaturated fatty acids, while gumminess was only correlated with the moisture content. The link between TPA and physico-chemical characteristics was established using multivariate matrix correlations and revealed that the pattern of texture attributes was mostly linked to the moisture and fat contents. Principal component analysis was carried out to identify the main physico-chemical properties and TPA attributes of the four classes of halwa as determined by cluster analysis. The four classes of halwa could be characterised as soft-resilient, soft-springy-cohesive, soft-springy and hard-chewy.

Keywords: food classification, fatty acids, sugar, multivariate analysis, cluster analysis, principal component analysis

(ProQuest: ... denotes formula omitted.)

INTRODUCTION

Halwa is a dessert (i.e., sweet jelly) made up mainly of starch, sugar, water, ghee, and flavoured with saffron, nuts and/or rose water. Omani halwa is usually made with sugar (50%), water (25%), mill flour (10%) and ghee (clarified butter) (15%) (AL-SHAMSI et al., 2013). Traditional halwa is prepared as follows. First, sugar is added to boiling water in an open pan. Starch is added to the sugar syrup after all of the sugar has dissolved. The heating process continues with manual mixing until the mixture turns into a thick, jelly-like substance. At this stage, fats, nuts, and spices are added, and then, the gels are poured into dishes of varying sizes for retail sale. There are mainly 3 types of Omani halwa, based on their final colour: white, yellow or black. They differ in the sweetener that is used, the flavour and texture. AL-SHAMSI et al (2013) used a structured survey questionnaire to determine consumer preferences and behaviour with regard to 22 attributes of halwa. Five types of consumer groups were identified through hierarchical cluster analysis (AL-SHAMSI et al, 2013). The type and amount of fat and sugar content in halwa can be considered part of an unhealthy diet. The successful development of healthier halwa (i.e., one that is lower in fat and sugar with the incorporation of health functionality) by the food industry requires an understanding of the different classes of halwa in the market and their generic attributes in order to explore the sensory preferences of consumers.

Product classification systems are considered a viable approach to support a specific surrogate. This system could be used as the basis for developing new products with varied specification or characteristics as well as for exploring the generic characteristics or attributes of low- or high-grade products to avoid adulteration. This classification could include general categories that can lead to specific relationships between product attributes (SOUSA and WALLACE, 2006). Qualitative information is difficult to use in high dimensions. PENNINGTON and FISHER (2009) pointed that the classification of fruits and vegetables is most helpful for dietary assessment if the classification is based on composition, as well as for teaching students about food composition and to assist dieticians in providing dietary guidance to patients and clients. They determined levels of food components in fruits and vegetables correlated with classification criteria based on botanical family, colour, the part of the plant, and the total antioxidant capacity. However, the level of total antioxidant capacity did not match well with the identified clusters. RELEAUX (1904) devised one of the first product classification systems after recognising the identicalness among various properties of different products. Products have been classified in several different ways. Different perspectives on what a product is and the purpose of the classification lead to distinct classification systems (SOUSA and WALLACE, 2006).

Food products can be classified by flavour compounds, chemical composition, and texture. MARTIN et al. (1999) classified vegetable oils by linear discriminant analysis of electronic nose data for flavour compounds. Classification studies have been performed on a variety of food products, such as food, beverages and perfumes (PENZA et al., 2001); commercial cheeses based on their volatile profiles using multivariate analysis of complex data of GC/ MS (DIRINCK and WINNE, 1999) and cheddar cheese based on flavour profiles (O'RIORDAN et al., 2003); Greek white wines from 6 grape varieties based on their amino acid profile (SOUFLEROS et al., 2003); apple fruit maturity according to polyphenol composition (ALONSOSALCES et al., 2005); medicinal plants of different species based on 10 triterpenoid acids (GUO et al., 2009); vegetable oils based on fatty acid profile (BRODNJAK-VONCINA et al., 2005); commercial potato chips based on colour and image texture (MENDOZA et al, 2007); and tomato fruit using acoustic impact and color - imetry (BALTAZAR et al, 2008). However, food products could be classified by their textural attributes. Instrumental TPA attributes were correlated with the composition of sandesh (a soft cheese made by direct acidification of heated milk with dilute acid followed by removal of whey and then mixed with sugar) available in various markets. The textural attributes were significantly influenced by the moisture, fat and ash contents (KHAMRUI and SOLANKI, 2010). There are few classifications of foods based on texture and composition.

The objective of this study was to classify Omani halwa available in the market based on chemical composition, physico-chemical properties and instrumental TPA attributes. This classification through a product survey could help the industry to understand existing products and to develop low-fat and low-sugar products (with new ingredients) by mimicking the TPA attributes of existing products.

MATERIALS AND METHODS

Halwa

Halwa samples (4 batches) were purchased from different local supermarkets in Muscat. The samples were selected from different types with various prices (very high, high, medium and low), colours (black and yellow) and flavours (royal, special, and saffron). After purchasing, the samples were stored at room temperature (20° C) for one day before analysis.

Proximate chemical analysis

Moisture, protein, fat, and ash were determined by following the AOAC Official Methods (2000). The moisture content was determined by drying 5 g samples in a mechanical convection oven for at least 18 hours at 105°C. Crude protein was determined by the Kjeldahl method using a conversion factor of 5.55 ? ?. Digestion and analysis were performed in a Digestion System 2000 and a Kjeldahl analyser unit 2300 (Foss Tecator, Höganäs, Sweden). The ash content of the samples was determined by burning the sample in a furnace in accordance with AOAC methods (2000). Approximately 5 g of each sample was weighed and burned at 530°C in a muffle furnace for 18 hours. The fat content was determined gravimetrically after extraction with light petroleum ether and evaporated at 103°C in an oven for at least 1 hour. The non-sucrose carbohydrate content was calculated by the difference of the total proximate composition and the water, protein, fat, ash and sucrose. All of the values were calculated in units of g/100 g sample. All of the analyses were performed in triplicate.

Sugars

Individual free sugars were determined by H PLC according to the technique of MYHARA et al. (1998). Sugars were extracted by refluxing halwa samples twice with aqueous ethanol (850 and 750 ml/1), and the residues were washed with ethanol (750 mL L^sup -1^). The extracts were then treated with lead acetate (10 g L^sup -1^) over gentle heat to precipitate proteins. After centrifugation, the supernatant solutions were dried in a rotary evaporator under reduced pressure. The dry residues were dissolved in a known volume of water, decolourised by passing through a C18 Sep-Pak solid-phase extraction tube (Waters Associates, Milford, MA, USA) and filtered through a 0.45 µp? cellulose acetate filter.

Sugars were separated on a Hewlett-Packard (Palo Alto, CA, USA) HP 1090 system, equipped with a 10 µL sample loop, a refractive index detector (model HP1047A) and a silica Nova-Pak Radial-Pak cartridge column (100 mm x 8 mm id, particle size 4 µm; Waters). Both the column and detector were maintained at 40°C. Prior to analysis, the silica column was modified by passing 1 L of acetonitrile/water (65:35) containing 1 g L^sup -1^ of 3-aminopropyltriethoxysilane at a flow rate of 1 mL min1. The column was then stabilised with 1 L of acetonitrile/water (65:35) containing 0. 1 g L^sup -1^ of 3-arninopropyltriethoxysilane. The latter solution was also used as the mobile phase. External standards of fructose, glucose and sucrose (filtered through C18Sep-Pak and 0.45 µm cellulose acetate filters) were used for quantification.

Colour

The colour of the halwa samples was determined using a colour meter at room temperature (Minolta CR-310, Minolta, Japan). The equipment was calibrated with a white standard calibration plate provided by the manufacturer. The halwa samples were placed on a flat plate, and the tip of the measuring head was pointed at the sample for measurement. Each value was the mean of 6 measurements. The results were expressed in Hunter L, a, b values, in which L is the lightness or darkness (black, L=0; white, L=100), +a is the redness, -a is the greenness, +b is the yellowness, and -b is the blueness. The colour (CL) was also calculated as CL= (L*b)/a (RAHMAN etat, 2002).

pH

The pH of the halwa solution was measured by preparing a 10% (w/v) halwa solution in distilled water (RAHMAN et al, 2008). The pH was then measured in triplicate with a glass electrode (Toledo MPC 227 pH meter, Mettler-Toledo GmbH, Schwerzenbach, Switzerland).

Fatty acids

Triplicate samples, each weighing approximately 2 g, were freeze-dried for 4 days. The fatty acids were extracted according to the method described by AOAC (2000) with some modifications. Halwa samples were mixed with 50 mL of 1% KOH, and 4 mL of tricosanoic acid internal standard (C23) was added. The mixture was heated for 30 min at 150°C, followed by cooling at room temperature. The samples were transferred to a 500 mL separating funnel, and 150 mL of distilled water was added followed by 0. 1% methyl orange drop-wise until the colour changed to yellow. A solution of 5 ? HCl was added until the colour turned light pink (pH=2), and the solution was then partitioned vigorously for 5 min with 100 mL of diethyl ether. The diethyl ether layer was collected, and the coloured aqueous phase was re-extracted with a further 100 mL of diethyl ether. The diethyl ether layer extracts were pooled and washed four times with 80 mL of distilled water and subsequently dried over anhydrous Na2S04. The diethyl ether extract was then concentrated to 2 mL in a rotary vacuum evaporator at less than 40°C and transferred to a screw-cap test tube following addition of 2 mL of BF3-methanol reagent (14%). The test tube was heated at 100°C for 15 min in a water bath. After cooling to room temperature, 3 mL of hexane and 5 mL of saturated NaCl were added, and the mixture was vigorously shaken for 5 min. The hexane layer was then carefully transferred using a Pasteur pipette into an amber crimp cap and stored at -20°C until assayed by gas chromatography.

Fatty acid samples were analysed with an Agilent Plus gas Chromatograph (GC) 6890N equipped with a 30 m × 0.25 mm SP-2380 (Supeclo Inc., USA) fused-capillary column attached to a flame ionisation detector. Helium was used as the carrier gas at a velocity of 20 cm/sec with electronic pressure control and split ratio of 50:1. The GC temperature was programmed from 50° to 250°C at a heating rate of 4°C/min. The temperatures of the injector and detector were 250° and 260°C, respectively. Fatty acids were identified by comparison of the retention times with a 37 FAME Standard mix solution (Supeclo). The concentration of individual fatty acids was calculated using tricosanoic acid (C23) as an internal standard.

Instrumental Texture Profile Analysis (TPA)

Instrumental TPA attributes were measured using a TA.XT2 Texture Profile Analyser (Stable Microsystems, Godalming, Surrey, UK) with two cycles of compression-decompression. The texture analyser was coupled with the computer software (texture expert). The load cell was calibrated with a 5 kg weight. The equipment was set to zero automatically before each experiment (i.e., the point when the top plate just touched the surface of the table). Compression mode was used for the textural analysis. Halwa samples were cut into 10, 20 and 30 mm cubes manually with a knife. The halwa specimens were placed on the stage of the texture analyser and compressed with a flat, 75 mm plunger (for 10, 20, and 30 mm cubes) or a flat, 20 mm, edge cylindrical plunger (for a 30 mm cube) or a plunger 5 mm in diameter (for a 10 mm cube). Halwa samples were compressed to 60% of their original height at a compression rate of 1.0 mm/ s at room temperature (i.e., 20°C). The force-displacement curve was automatically recorded by a computer attached to the TPA. At least six replicates were analysed for each halwa sample.

The TPA attributes were determined from force-time curves based on the following attributes: hardness (HA), the maximum force during the first compression cycle (N); adhesiveness (AD), the negative area of the first compression (N m); springiness 1 (SPI), the distance between the end of the first compression and start of the second (mm); springiness 2 (SP2), the distance between the start of the first compression and the starting point; firmness 1 (FRI), the slope of the initial linear part of the first compression (N/s); firmness 2 (FR2), the slope of the second linear part of the first compression (N/s); cohesiveness 1 (COI), the ratio of the positive area of the second compression and the positive area of the first compression; cohesiveness 2 (C02), the ratio of difference of the positive and negative areas of the second compression and difference of the positive and negative areas of the first compression; resilience (RE), the ratio of the negative area to positive area of the first compression; gumminess 1 (GUI), the product of hardness and cohesiveness 1 (N); gumminess 2 (GU2), the product of hardness and cohesiveness 2 (N); chewiness 1 (CHI), the product of gumminess 1 and springiness 1 (N m); and chewiness 2 (CH2), the product of gumminess 2 and springiness 1 (N m) (RAHMAN and AL-MAHROUQI, 2009).

Multivariate analysis

Multivariate analysis (Spearman's correlation matrix, cluster analysis, and principal component analysis) was used to identify different groups of halwa and their characteristics based on instrumental texture and physico-chemical properties. These analyses were performed with PAST software (HAMMER et ed., 2001). Correlations were used to determine the relationships between each variable. The Spearman's correlation matrix of the TPA attributes and the physico-chemical variables were detemiined, and the results were summarised for the ? values < 0.05 and 0.01. A cluster analysis was performed to group similar halwa. A hierarchical clustering using Ward's method of variance minimisation within groups was applied to the TPA standardised attributes and the physico-chemical variables (WARD, 1963). The number and composition of the clusters retained were estimated by visual observation of the clustering tree. The multivariate linkage between the overall texture and the overall physico-chemical composition was investigated using the Bio-Env procedure of "Primer" (CLARKE and GORLEY, 2006). This procedure measures the correlations between a single matrix of similarities/distances calculated from the texture profile and all of the matrices of similarities calculated from all of the possible combinations of physico-chemical composition variables [(2n - 1) matrices, where ? is the number of variable]. The best correlation identifies the best linkage between the texture profile matrix and one of the 2n - 1 physico-chemical composition matrices. The better correlation matrices are based on "ranks of values" in the two similarity matrices.

The Spearman's and Pearson's correlations were plotted to determine the degree of non-linearity of the variables. Principal component analysis was used to classify and to characterise the products. In the case of principal component and cluster analysis, the TPA attributes and physico-chemical properties were standardised to remove undesired effects related to the size of the measurement by the following equation:

... (1)

where y is the standardised data (or variable), x is a variable, is the average, and s is the standard deviation of the data for this variable.

RESULTS AND DISCUSSION

In this study, batch 1 of the halwa was used to check the reproducibility of instrumental TPA data generated by compression and punching methods. Table 1 shows the TPA attributes measured by punching with a 20 mm probe provided more reproducible data compared with the 75 mm probe compression and punching with a 5 mm probe. For example, the standard deviations of hardness compared with the average values were 88.5, 21.4 and 6.3% for the 5 mm probe puncher, the 75 mm compression disk (sample size of 30 mm) and the 20 mm probe puncher methods, respectively. To determine the sensitivity of the 20 mm punch-probe method, batch 2 with varying moisture contents of halwa were used. The TPA attributes using the 20 mm probe did not show much sensitivity (Table 2) while moisture varied from 13. 1 to 18.9 g/ 100 g halwa (p>0. 19) (Table 3). However, the compression disk method was found to be more sensitive as a function of sample size and moisture content (Tables 1 and 4). Thus, the compression method was used to determine the instrumental TPA attributes for a 10 mm cube of halwa (the most reproducible sample size). The variability of the TPA data using different probes was reported. The sensitivity results of 3 types of punching probes (blade, sphere and cylinder) indicated that cylindrical probe produced the most sensitivity with respect to the TPA attributes of salmon fillet (CASAS et ed., 2006). Similarly, BENEDITO et cd. (2006) compared the instrumental attributes of cheese as a function of its maturity using different types of probes (compression, punching and needle). They observed a similar correlation in the cases of compression and punching, which was much higher compared to the needle-type probe.

Batch 3 was purchased and included 15 samples (black and yellow halwa) from different shops. The TPA attributes were determined from the typical, instrumental TPA force-time graph using a 75 mm disk (Fig. 1), which shows compression-decompression cycles. The instrumental TPA attributes of 1 5 samples are shown in Table 4, and their chemical compositions are given in Table 5. The extraction of fat through Gas Chromatography was considered for classification. Hardness, adhesiveness, firmness, cohesiveness, resilience, gumminess, and chewiness decreased with an increase in the water content, while springiness increased with an increase in the moisture content (Table 4). Table 5 shows detailed chemical compositions of 15 commercial products. Moisture, protein, fat, ash, sucrose, and pH varied from 11.2-19.9, 0.15-1.10, 4.313.5, 0.10-0.84, 14.9-56.3 g/100 g halwa, and 4.26-5.46, respectively. It is interesting to see that most of the black halwa contained more ash compared with the yellow halwa. This greater amount of ash could have resulted from the use of brown or raw sugar in preparing black halwa whereas yellow halwa was prepared with white sugar. Table 6 shows the significant relationship between the variables at p<0.05 and p<0.0 1 . For example, hardness was significantly correlated with AD, SPI and SP2, GUI and GU2, XW, SU, and C20:2 and C20:3n6 fatty acids (p<0.05). Moisture, the types of fat and sugar significantly affected most of the TPA attributes (Table 6). In the case of ready-to-eat meat products, TPA attributes were significantly correlated with the water and fat contents (PROBOLA and ZANDER, 2007). The textural attribute matrix found using the Bio-Env procedure showed the highest correlation with the moisture and unsaturated fat contents followed by total fat and non-sugar, and sucrose and saturated fat (Table 7).

Hardness was negatively correlated with moisture content (p<0.05). Similarly, cheese with resistance starch showed that hardness decreased with increasing water content using a 35 mm disk (NORONHA et al, 2007). In food products, water acts as plasticiser of the solid constituents, thus making the food less elastic and more susceptible to fracture upon compression (FOX et al., 2000). Water molecules are bound within a three-dimensional matrix and weaken the structure of the network. Thus, increasing the moisture content reduced the coherence of the matrix and resulted in softer products (DIMITRELI and THOMAREIS, 2007). In the case of compression of cheese using a 35 mm disk, the hardness increased with fat content when most of the composition remained the same, except for the pH and butanoic acid content (NORONHA et al, 2008). Similarly, VOLIKAKIS et al (2004) observed that the hardness increased with increasing fat content. However, DIMITRELI and THOMAREIS (2007) observed a decrease in hardness with increasing fat content. Fat could behave as an active filler, and thus, hardness increased with increasing fat content. However, if fat acts as a lubricant, the hardness could decrease with fat content (VOLIKAKIS et al., 2004). MUGUERZA et al. (2002) observed increasing hardness with decreasing fat content in the case of meat sausage. In the case of sausage, hardness was significantly correlated with water but not with fat content (HERRERO et al., 2008). In the case of dairy products such as sandesh, hardness was positively (p<0.0 1) correlated with fat content but negatively (p<0.0 1) correlated with the moisture and ash contents and did not show any significant effect from the sugar and protein content or acidity (KHAMRUI and SOLANKI, 2010).

Adhesiveness was significantly correlated with moisture, sucrose, non-sucrose starch, and some fatty acids (CI 7:0, CI 8:0, CI 5:1, C18:ln9t, C18:ln9c, C20:2, C20:3n6, C20:3n3) (p<0.05). For the AD of milk-based sweets determined compression with a 75 mm disk, the moisture showed negative correlations while fat showed positive correlations (p<0.01) (KHAMRUI and SOLANKI, 2010). Springiness 1 and springiness 2 were significantly correlated with sucrose, non-sucrose and some fatty acids (C17:0, C18:0, C14:l, C15:l, C18:ln9t, C18:ln9c, C18:2n6t, C20:l, C20:2, C20:3n6, C20:3n3) (p<0.05). In the case of compression of cheese with a 75 mm disk, VOLIKAKIS et al. (2004) observed no significant effect of fat on springiness 1. The SPI of milk-based sweets showed positive and negative correlations (p<0.05) with the fat and moisture contents (KHAMRUI and SOLANKI, 2010). Firmness 1 was significantly correlated with the water, total fat, saturated fat and unsaturated fat contents, and fatty acids (CI 6:0, CI 5:1, C18:ln9t, C18:ln9c) (p<0.05). Firmness 2 was as significantly correlated as firmness 1 with the exception of fatty acids CI 5:1 and C18:ln9c.

Cohesiveness 1 was significantly correlated with the protein, total fat, saturated and unsaturated fat contents and fatty acids (C16:0, C17:0, C18:0, C15:l, C18:ln9t, C18:ln9c, C20:2, C20:3n6 and C20:3n3) (p<0.05). Similarly, cohesiveness 2 was significantly correlated to the same variables as cohesiveness 1 except for protein and with the addition of sucrose. In the case of cheese, cohesiveness 1 did not show any specific trends with fat content when most of the composition remained the same except for pH and butanoic acid content (NORONHA et al., 2008). Similarly, VOLIKAKIS et al. (2004) observed that cohesiveness 1 did not show any trend with increasing fat content. In the case of sausage, cohesiveness 1, adhesiveness and springiness 1 did not show any correlation with the water or fat contents (p<0.05) (HERRERO et al., 2008). Cohesiveness 1 and springiness 1 of ready- to-eat meat products were correlated (i.e., rvalues) with water, fat and protein as follows: 0.25, -0.21, -0.38 and 0.50, -0.54, -0.38, respectively (PROBOLA and ZANDER, 2007). In the case of dry fermented sausage (i.e., considering water, fat, water activity and pH), HA did not correlate with any parameters (p>0.05), COI correlated with water content (r2 = 0.306, p<0.005), AD correlated with water content and pH (r2 = 0.282, p<0.005), and SP correlated with only water content (r2 = 0.2 16, p<0.05) (HERRERO et al., 2008). COI of milk-based sweets was not significantly correlated with the composition parameters (water, fat, protein, ash, sugar contents and acidity) (p>0.05).

Resilience was significantly correlated with the total fat, saturated and unsaturated fat contents and fatty acids (C4:0, C16:0, C17:0, C18:0, C14:l, C15:l, C18:ln9t, C18:ln9c, C20:2, C20:3n6) (p<0.05). In the case of dairy sweets, RE was correlated positively (p<0.01) with the ash and water contents, whereas the fat content negatively influenced RE (KHAMRUI and SOLANKI, 2010). Gumminess 1 and gumminess 2 were significantly correlated with the water content and CI 6:0 fatty acids (p<0.05). In the case of high and low fat cheese punctured with a 1 0 mm probe, firmness 1, cohesiveness 1, and gumminess 1 significantly increased with an increase in water (SAINT-EVE et al, 2009). In the case of dairy products such as sandesh, gumminess 1 was positively (p<0.01) correlated with the fat content but negatively (p<0.01) correlated with the moisture and ash contents and did not show any significant effect from the sugar and protein content or acidity (KHAMRUI and SOLANKI, 2010).

Chewiness 1 and chewiness 2 were significantly correlated with water, total and unsaturated fat, and fatty acids (C16:0, C18:ln9t, C18:ln9c) (p<0.05). In the case of cheese, gumminess 1 and chewiness 2 did not show any trend with fat content (VOLIKAKIS et al., 2004). CHI of milk-based sweets was negatively correlated with water content and was positively influenced by the fat content (p<0.01). The salt content also affected cohesiveness 1 differently in low-fat and high-fat cheeses. The cohesiveness 1 was only significantly affected by the fat content at low salt levels but not at high salt levels. The TPA parameters (FAI, AD, COI, SPI, and GUI) differentiated the products according to salt content rather than at low-fat content. However, fat played a major role in aroma release and olfactory perception, whereas salt played a predominant role in sensory texture and TPA parameters (SAINT-EVE et al, 2009). Considering all of the physico-chemical properties and TPA attributes, a cluster analysis based on Ward's method revealed 4 distinct groups of formulation, named Gl, G2, G3, and G4 at a level of similarity/distance of 9 (Fig. 2).

The plot of Pearson's versus Spearman's correlations suggests no real bias and supports the idea of linkages between variables being mostly linear (Fig. 3). The PCA analysis shows that seven principal components (93% of the total variance) had eigenvalues close to 1 (Kaiser criterion; RAHMAN and AFFARSI, 2005). These components explained 41, 19, 15, 8, 4, 3 and 3% of the total variance (Table 8). The first axis was highly correlated with firmness 1 , firmness 2, cohesiveness 1, cohesiveness 2, total fat, saturated and unsaturated fat and fatty acids (C4:0, C14:0, C15:0, C16:0, C17:0, C18:0, C14:l, C15:l, C18:ln9t, C18:2n6t) and negatively correlated with adhesiveness, resilience and fatty acids (C18:ln9c, C20:2, C20:3n6) (Table 8) and corresponds to a descriptor of plasticity (deformation and structural damage of first and second compression). The second axis was correlated to firmness 2, cohesiveness 1, chewiness 1, chewiness 2, protein and fatty acids (C8:0, C10:0, C12:0, C18:2n6c) and corresponds to the elasticity (force and deformation of first and second compression). The third axis was strongly correlated with hardness and moisture content as well as adhesiveness, springiness 1 , springiness 2, gummlness 1, gumminess 2 with sugar content and corresponds to the hardness (force needed to deform the product). The fourth axis correlated with springiness 2, chewiness 1, chewiness 2 and corresponds mainly to chewiness (the work required to chew the product for swallowing), while the fifth axis correlated with protein, colour and CI 6:1 fatty acid and corresponds mainly to protein and colour. The sixth axis correlated with sucrose, non-sucrose and pH, which correspond to the sugar and carbohydrate content. The seventh axis is correlated with colour and C20:2 fatty acids, thus corresponding to the colour. In general, specific fatty acids contribute to most of the principal components.

The four identified groups by cluster analysis are shown on a PCA biplot with the projection of the original TPA variables (Fig. 4). Figure 4 shows that group Gl (on the left-hand side of the PCA plot) includes yellowish and black halwa leading to soft and resilient attributes (low-fat and high sugar). On the opposite direction to that of the resilient component, group G2 includes halwa that are soft and cohesive-springy (high-fat and low sugar). G4 includes soft and springy halwa (high-fat and medium-sugar). This result indicates that most of the products in the market are soft with different textural attributes. Group G3 opposite to the G4 corresponds to hard and chewy (fatty and medium sugar) halwa. The springiness of the softer products indicated that structural damage due to compression was rapidly repaired. SOUKOULIS et cd. (2008) identified two classes of ice cream formulated with different hydrocolloids. LASSOUED et al (2008) classified 15 baked products (i.e., bread) with various types of dough similar to commercial products, and they classified the products based on sensory texture and image analysis. They grouped the products into 6 clusters based on Hierarchical Cluster Analysis and identified suffer products (i.e., stiffness, elasticity and flexibility) (group 1), products with largercrumb cells (group 2), elastic products (i.e., elasticity, softness, crumbliness) (group 3), products with fine-crumb cells (group 4), products with cell regularity (group 5), and crumble-bread that was easy to tear (group 6). In the case of 12 commercial custard desserts, DE WIJK et al. (2003) identified 4 cluster groups based on sensory and instrumental mouth feel. They used PCA and cluster analysis and characterised the classes by roughness, thickness, melting, creaminess/ softness. The halwa industry could produce their products based on these categories in order to imitate the types of products available in the market today and possibly to please the consumer (AL-SHAMSI et al., 20 13).

CONCLUSION

Commercial halwa products purchased in the market were surveyed and classified by measuring their physico-chemical properties and TPA attributes. The instrumental TPA attributes of halwa were correlated with the moisture, fat, fatty acid and sugar contents. Based on the physico-chemical properties and TPA attributes, halwa in the local market could be classified into 4 groups: hard-chewy (fatty and medium-sugar), soft-resilient (low-fat and highsugar), soft-springy (fatty and medium-sugar) and soft-cohesive-springy (high-fatty and low sugar). PCA was used to identify 7 principal components, which correspond to the plasticity, elasticity, hardness, chewiness, proteincolour, sugar-carbohydrate, and colour. Halwa producers could develop new products based on the characteristics of these four classes in order to please the consumer base.

References

REFERENCES

Al-Shamsi Q., Rahman M.S., Abdullah A. and Claereboudt M.R. 2013. Consummers' preference mapping and their behavior with regard to traditional halwa using a structured survey questionnaire. Journal of Food Products Marketing. In Press.

Alonso-Salces R.M., Herrero G, Barranco ?., Berrueta L.A., Gallo ?. and Vicente, F. 2005. Classification of apple fruits according to their maturity state by the pattern recognition analysis of their poryphenolic compositions. Food Chemistry. 93(1): 113-123.

AOAC. 1990. Official methods of analysis. Association of Official Analytical Chemists. 15th Edition. Virginia, USA.

Balabin R.M. and Safieva R.Z. 2011. Biodiesel classification by base stock type (vegetable oil) using near infrared spectroscopy data. Analytica Chimica Acta. 689(2): 190-197.

Baltazar ?., Aranda J.I. and Gonzalez -Aguilar G. 2008. Bayesian classification of ripening stages of tomato fruit using acoustic impact and colorimeter sensor data. Computers and Electronics in Agriculture. 60(2): 113-121.

Benedite J., Simal S., Clemente G. and Mulet ?. 2006. Manchego cheese texture evaluation by ultrasonic and surface probes. International Dairy Journal. 16: 431-438.

Brodnjak-Voncina D., Kodba C.Z. and Novic M. 2005. Multivariate data analysis in classification of vegetable oils characterized by the content of fatty acids. Chemometrics and Intelligent Laboratory Systems. 75: 31-43.

Casas C, Martinez O., Guillen M.D., Pin C. and Salmerón J. 2006. Textural properties of raw Atlantic salmon (Salmo Salar) at three points along the fillet, determined by different methods. Food Control. 17: 511-515.

Clarke K.R. and Gorley R.N. 2006. PRIMER v6: User Manual/Tutorial. PRIMER-?, Plymouth.

De Wyk R.A., Gemert L.J., Terpstra M.E.J, and Wilkinson C.L. 2003. Texture of semi-solids; sensory and instrumental measurements on vanilla custard desserts. Food Quality and Preference. 14(4): 305-317.

Dimitreli G. and Thomareis A.S. 2007. Texture evaluation of block-type processed cheese as a function of chemical composition and in relation to its apparent viscosity. Journal of Food Engineering. 79(4): 1364-1373.

Dirinck P. and De Winne A. 1999. Flavor characterization and classification of cheeses by gas chromatographicmass spectrometric profiling. Journal of Chromatography A. 847: 203-208.

Eymery O. and Pangborn R.M. 1988. Influence of fat, citric acid and sodium chloride on texture and taste of a cheese analog. Science des Aliments. 8: 15-32.

Fox P.F., Guinee T.P., Cogan T.M. and McSweeney P.L.H. 2000. Fundamentals of cheese science, Aspen Publishers, Inc., Gaithersburg, Maryland.

Guo S., Duan J., Tang Y., Su S., Shang E., Ni S. and Qian D. 2009. High-performance liquid chromatography-Two wavelength detection of triterpenoid acids from the fruits of Ziziphus Jujuba containing various cultivars in different regions and classification using chemometric analysis. Journal of Pharmaceutical and Biomedical Analysis. 49: 1296-1302.

Herrero A.M., Hoz L., Ordonez J.A., Herranz B., Romero M.D.R. and Camberò M.I. 2008. Tensile properties of cooked meat sausages and their correlation with texture profile analysis (TPA) parameters and physico-chemical characteristics. Meat Science. 80(3): 690-696.

Herrero A.M., Ordonez J.A., Avila R., Herranz B., Hoz L. and Camberò M.I. 2007. Breaking strength of dry fermented sausages and their correlation with texture profile analysis (TPA) and physico-chemical characteristics. Meat Science. 77(3): 331-338.

Khamrui K. and Solanki D.C. 2010. The relationship of textural characteristics with composition of sandesh produced from various market ntilk classes. International Journal of Dairy Technology. 63(3): 451-456.

Lassoued N., Delarue J., Launay B. and Michon C. 2008. Baked product texture: Correlations between instrumental and sensory characterization using Flash Profile. Journal of Cereal Science. 48(1): 133-143.

Martin Y. G., Pavon J.L.P., Cordero B.M. and Pinto CG. 1999. Classification of vegetable oils by linear discriminant analysis of Electronic Nose data. Analytica Chimica Acta. 384: 83-94.

Mendoza F., Dejmek P. and Aguilera J.M. 2007. Color and image texture analysis in classification of commercial potato chips. Food Research International. 40(9): 1146-1154.

Muguerza E., Fista G., Ansorena D., Astiasaran I. and Bloukas J.G. 2002. Effect of fat level and partial replacement of pork backfat with olive oil on processing and quality characteristics of fermented sausages. Meat Science. 61(4): 397-404.

Myhara R.M., Taylor M.S., Slominski B. and Al-Bulushi I. 1998. Moisture sorption isotherms and chemical composition of Omani dates. Journal of Food Engineering. 37: 471-479.

Noronha N., Cronin D., O'Riordan D. and O'Sullivan M. 2008. Flavoring reduced fat high fiber cheese products with enzyme modified cheeses (EMCs). Food Chemistry. 110(4): 973-978.

Noronha N., O'Riordan E.D. and O'Sullivan M. 2007. Replacement of fat with functional fiber in imitation cheese. International Dairy Journal. 17(9): 1073-1082.

O'Riordan P.J. and Delahunty CM. 2003. Characterization of commercial Cheddar cheese flavor. 2: Study of Cheddar cheese discrimination by composition, volatile compounds and descriptive flavor assessment. International Dairy Journal. 13: 371-389.

Pennington J.A.T. and Fisher RA. 2009. Classification of fruits and vegetables. Journal of Food Composition and Analysis. 22(1): S23-S31.

Penza M., Cassano G., Tortorella F. and Zaccaria G. 2001. Classification of food, beverages and perfumes by W03 thin-film sensors array and pattern recognition techniques. Sensors and Actuators B. 73: 76-87.

Probola G. and Zander L. 2007. Application of PCA method for characterization of textural properties of selected ready-to-eat meat products. Journal of Food Engineering. 83(1): 93-98.

Rahman M.S. and ??-Farsi S. 2005. Instrumental texture profile analysis (TPA) of date flesh as a function of moisture content. Journal of Food Engineering. 66: 505-51 1.

Rahman M.S., Al-Amri O. and Al-Bulushi I.M. 2002. Pores and physico-chemical characteristics of dried tuna produced by different methods of drying. Journal of Food Engineering. 53: 301-313.

Rahman M.S. and Al-Mahrouqi A.I. 2009. Instrumental texture profile analysis of gelatin gel extracted from grouper skin and commercial (bovine and porcine) gelatin gels. International Journal of Food Sciences and Nutrition. 60(S7): 229-242.

Rahman M.S., Al-Saidi G.S. and Guizani N. 2008. Thermal characterization of gelatin extracted from yellowfin tuna skin and commercial mammalian gelatin. Food Chemistry. 108: 472-481.

Saint-Eve ?., Lauverjat C, Magnan C, Deleris I. and Souchon I. 2009. Reducing salt and fat content: Impact of composition, texture and cognitive interactions on the perception of flavored model cheese. Food Chemistry. 116: 167-175.

Siripatrawan U. 2008. Self-organizing algorithm for classification of packaged fresh vegetable potentially contaminated with foodborne pathogens. Sensors and Actuators B: Chemical. 128(2): 435-441.

Soufleros E.H., Bouloumpasi E., Tsarchopoulos C. and Biliaderis CG. 2003. Primary amino acid profiles of Greek white wines and their use in classification according to variety, origin and vintage. Food Chemistry. 80: 261-273.

Soukoulis C, Chandrinos L. and Tzia C 2008. Study of the functionality of selected hydrocolloids and their blends with K-carrageenan on storage quality of vanilla ice cream. LWT- Food Science and Technology. 41(10): 1816-1827.

Sousa I. and Wallace D. 2006. Product classification to support approximate life-cycle assessment of design concepts. Technological Forecasting and Social Change. 73(3): 228-249.

Volikakis P., Biliaderis CG., Vamvakas C and Zerfiridis G.K. 2004. Effects of a commercial oat-ß-glucan concentrate on the chemical, physico-chemical and sensory attributes of a low-fat white-brined cheese product. Food Research International. 37(1): 83-94.

Paper received November 15, Accepted February 7, 2012