Differentiation of ripe banana flour using mineral composition and logistic regression model

Cavendish (Musa paradisiaca L, cv cavendshii) and Dream (Musa acuminata colla. AAA, cv ‘Berangan’) banana flours were prepared from ripe fruits collected from eleven markets located in Penang, Malaysia. The mineral composition (Na, K, Ca, Mg, Cu, Fe, Mn, Zn) of the flour were analyzed by atomic absorption spectrophotometer and the data obtained were analyzed using logistic regression model. Ripe banana flours were rich source of K and a fair source of other minerals, however logistic regression model identified Mg as an indicator to discriminate between the two types of banana flour affording 100 % correct assignation. Based on this result, mineral analysis may be suggested as a method to authenticate ripe banana flour. This study also presents the usefulness of logistic regression technique for analysis and interpretation of complex data. Keyword: Mineral composition, logistic regression model, ripe banana flour 84 Abbas F. M. A., Saifullah, R. and Azhar, M. E. International Food Research Journal 16: 83-87 Materials and methods Preparation of the sample Two of the most common banana varieties, namely Cavendish (Musa paradisiaca L, cv cavendshii) and Dream (Musa acuminata colla. AAA, cv ‘Berangan’) banana, were purchased from eleven markets around Penang, Malaysia. A total of 70-80 ripe banana of each variety were obtained from each market (A total of 1648 bananas for all samples). The fruits were peeled and cut into transverse slices of about 2 mm thickness. To reduce enzymic browning, slices were then dipped in 0.5% (w/v) sodium metabisulphite solution for 5 min, drained and dried in oven (AFOS Mini Kiln, at 60oC overnight). The dried samples were ground in a Retsch Mill Laboratory (Retsch AS200) to pass through 60 mesh screen to obtain banana flour. The flour was stored in airtight plastic packs in cold storage (15±2°C) for further analyses. Proximate composition Proximate analysis of banana flour was performed in triplicate following AOAC (1995) procedures and included the following : moisture by vacuum oven (method 934.06), protein by Kjeldahl nitrogen (method 920.152), and ash by direct analysis (method 940.26). The percentage of crude protein was estimated by multiplying the total nitrogen content by a factor of 6.25 (AOAC 1995). The lipid content was determined according to Bligh and Dyer (1959). Total carbohydrates were calculated by subtracting the total percent values of other measurements from 100. Proximate analyses were expressed as grams per 100 g of flour. Analysis of minerals For digestion, Millestone Ethos 900 microwave closed system was used. The flour samples were accurately weighed at 1.00 g and were transferred to 100 ml teflon vessels. Samples were digested with 6 ml of HNO3 (65%), 1 ml of H2O2 (30%) in the microwave digestion system for 31 min (digestion conditions for microwave were: 2 min at 250 W, 2 min at 0 W, 6 min at 250 W, 5 min at 400 W, 8 min at 550 W and vent for 8 min). The digest for Ca and Mg were added with 0.5% (w/v) Lanthanum solution, otherwise for Na (+ 1 mg K+/ml) and K (+ 1 mg N+/ml) respectively, and final diluted to 100 ml with De-ionised water (18.2 MΩ cm). Blank digestion was carried out in the same way. For the elemental analysis, a Perkin-Elmer Analyst 3110 atomic absorption spectrometer with hollow cathode lamps was used in this study. All measurements were carried out in an air/acetylene flame. Statistical methods Logistic regression model Logistic regression model or logit deals with the binary case, where the response variable consists of just two categorical values. Logistic regression model is mainly used to identify the relationship between one or more explanatory variables i X and the dependent variableY . Logistic regression model has been used for prediction and determining the most influential explanatory variables on the dependent variable (Cox and Snell, 1994). The logistic regression model for the dependence of i p (response probability) on the values of k explanatory variables k x x x ..., , , 2 1 is given in below (Collett, 2003): ki k i i i i x x p p P it b b b + + + = − = ... ) 1 ( log ) ( log 1 1 0 (1) Or ) ... ( exp 1 ... ( exp