An Assessment of the Women of Project Have Hope And Conditions Relating to Chronic Nutritional Health Problems Kristen Heimall 12/13/2009 CS 105 Introduction Project Have Hope is a nonprofit that works with 100 women living in the Acholi Quarter of Uganda. As most of the residents have been displaced from Northern Uganda due to civil war, there is very little data concerning the demographics of the population. While working as a volunteer, I constructed a survey to assess the demographic make-up, housing, education, nutrition levels, and interests of the women and their families. Through the use of data mining techniques, specifically classification learning, I developed a model that would predict whether the respondent and/ or one of their family members is suffering or has ever suffered from a chronic condition due to poor nutrition based on the other information gathered by the survey. In addition, I created two data graphics to display additional aspects of the dataset that were not considered in the model. Dataset Description The dataset used for this project was compiled from 97 respondents of the survey which assessed the following information: age, marital status, number of children, ownership of land, literacy, vocational training, child literacy, education, nutrition levels, typical day’s meals, chronic conditions/ death associated with poor nutrition and impact of Project Have Hope. Below is a more detailed description of the attributes. Attribute Name Type and Description Age Numeric; age of survey respondent Marital Status Nominal; marital status of respondent, either Married, Divorced, Single, or Widowed Number of Birth Children Numeric; number of biological children of respondent Number of Children Living with Individual Numeric; number of children, both biological and non-biological living with respondent Number of Children Cared for Numeric; number of children under care of respondent but do not currently live with respondent Own Land Nominal; whether respondent or husband has ownership of land, either yes or no Literate Nominal; whether respondent is literate, either yes or no Formal Education Nominal; whether respondent has received formal education, either yes or no Taken Vocation Training Nominal; whether respondent has taken part in thevocational training offered, either yes or no Interested in Vocation Training Nominal; whether the respondent has an interest is taking vocational training; either yes or no Number of Literate Children Numeric; number of biological and non-biological children of respondent who are literate Number of Children Enrolled in School Numeric; number of biological and non-biological children of respondent who are currently enrolled in school Severity of Problems with Nutrition Nominal; respondent rates the severity of their problems with nutrition on a one to five scale Meals per Day Numeric; number of meals respondent eats per day How often do you consume fruit/ vegetables Nominal; how often respondent consumes fruit/vegetables, either Never, Once in a while, Once a month, Once every two weeks, Once a week, Twice a week, Three times a week, Four times a week, Once a day, Twice a day, Three times a day, or Four times a day How often do you consume meat Nominal; how often respondent consumes meat, either Never, Once a year, Once every two months, Once a month, Twice a month, Three times a month, Once every two weeks, Once a week, Twice a week, Once a day, or Four times a day Chronic condition associated with poor nutrition Nominal; whether the respondent or a family member has ever suffered from a chronic condition due to poor nutrition, either yes or no Lost a family member because of poor nutrition Nominal, whether a family member of the respondent has ever died because of poor nutrition Rate how much PHH has benefited you Nominal; respondent rates how much Project Have Hope has benefited them on a one to five scale Data Preparation Since the data was originally 97 separate surveys, the first step in data preparation that I had to perform was to compile the answers onto an excel spreadsheet. After eliminating certain attributes that would have been inappropriate to include (for instance I removed the name attribute as that would have been chosen by the algorithm as the best indicator to predict the class), I converted the spreadsheet to a CSV (comma-deliminated) file. Once uploaded into Weka and saved as a .arff file, two attributes, Severity of problems with nutrition and Rate how much PHH has benefited you, had to be manually changed to nominal attributes by inputting the set of possible values for each. I then used Weka to randomize and split the dataset into the training data that would be used to develop the model, and test data which would be used to test the model. I used approximately 80% of the data (or 77 instances) as the training data and 20% (or 20 instances) as the test data.Data Analysis The data-mining algorithms I chose to use on this dataset are the 1R algorithm and the J48 tree algorithm. The 1R algorithm predicts the class based on one rule formulated from another attribute which produces the lowest error rate. The J48 tree algorithm constructs a decision tree based on a process similar to 1R, although the J48 algorithm looks at more than one attribute by recursively performing the process at each decision node. It also “prunes” the tree if necessary to ensure that it best generalizes the data. Results The model produced from the training data using the 1R algorithm had an accuracy rate of approximately 77%. It chose to use the “How often do you consume fruit/ vegetables” attribute as the best predictor of whether the respondent and/ or her family has ever suffered from a chronic condition due to poor nutrition. Below are the model and its corresponding confusion matrix. Rules based on 1R algorithm: Confusion Matrix: How often do you consume fruits/ veg: Once in a while -> No Predicted Class Once a week -> No No Yes Once a day -> No Actual Class Once a month -> No No 48 2 15 9 Three times a week -> No Four times a week -> Yes Yes Never ->