A STUDY ON FRAGILE STATE INDEX USING LISPMINER AND WEKA

I performed the discretization on the decision feature as many null values were present. In this project, I have added 6 new features along with the 12 indicators of FSI. The data for these new features have been taken from the World Bank website. I have selected these new features based on the weight they add to the existing data. Very critical parameters have been intensely studied and chosen to determine the best precision.I perfomed below data cleaning and pre-processing tasks with the use of WEKA.

• 1. Removal of Outlier
• 2. Removal of special characters so that csv files can be imported successfully in weka.
• 3. Fill up the missing values by the mean value of that column.
• 4. Deleted the empty rows from the data so that analysis can be done on the data.
• 5. Converted each numeric column value in the range of 0 to 10.
• 6. Converted the partitioning attributed to discretized or nominal attributes.

After completion of data pre-processing, I tried and tested a lot of classifiers in WEKA tool to get the best-suited classifier for our data. Few of the tested classifiers are as follows:

• 1. Logistic Model Tree (LMT),
• 2. Random Tree,
• 3. Random forest(bootstrapping algorithm) with decision tree model,
• Hoeffding Tree

I have also used LISp Miner to find out Action rules.LISp-Miner is an academic project for support research and teaching of knowledge discovery in databases. In order to find the action rules antecedent variable and stable variables were chosen.Also, finalized the succedent variable and stable attributes.The decision attribute was chosen from the given data with values 'Alert' -> ‘Stable’ and 'Alert' -> 'Warning' as succedent variable. Rules were chosen with highest confidence were chosen which had the range from 0.8 to 1.00.
This action rules suggests what changes in values of classification features are needed to lower FSI.