Test scaling Logistic regression (ML classification algorithm) by partitioning data by Histogram [2]

Information about important classes..

Class	Description
GetMinMax.java	Calculate the min max values of each feature
HistogramEnsembler.java	Encapsulate partitioning data through creation of models for each group. Also the model ensembling logic is implemented.
HistogramHelper.java	This calculates the bin number and converting a given coordinate to a bin number.
HistogramTester.java	Analysis of the partitioned data using a Histogram
HistogramTree.java	Allow formulating and manipulation of the histogram.
IHistogramHelper.java	Defines the contract of the histogram.
LogisticRegresionTester.java	Analysis of the logistic regression with full data set.
Metrics.java	Efficiency calculation
RandomPartitionedEnSembler.java	Encapsulate random partitioning data through creation of models for each group. Also the model ensembling logic is implemented.
RandomPartitionTester.java	Analysis of the logistic regression with randomly partitioned data.
TestHistogramHelper.java	This does the unit tests of the HistogramHelper class. It tests the dimensions and the change axis order.
TestHistogramTree.java	This does the unit tests of the HistogramTree class. It tests the neighbouring bins and checks the grops are of equal sizes.

Testing and Evaluation

	Technique Used	Accuracy (%)	Performance
Method 01	Logistic regression with full data set	74.2720201581548	training time : 52s Prediction time: 12s
Method 02	Logistic regression with randomly partitioned data set	74.35145399566024	training time: 2min42s Prediction time: 15s
Method 03	Logistic regression with partitioned data using histogram.	74.36789359939304	training time: 06min07s Prediction time: 12s

Above table is being used to compare three different methods used to run logistic regression.

According to the statistics given in the table in terms of accuracy all three methods shows relatively similar accuracy levels. When comparing training times method 01 shows the best time of 52s whereas method 03 shows worst. however considering the prediction times method 01 and method 03 shows best of 12s.