使用Weka提高正确分类实例的百分比

Question

我一直在将J48分类器用于具有1,000,000条记录和25,011条记录的训练数据的测试数据。我的输出中正确分类实例的数量非常少（3.4％）。我不确定如何改善这些数字。

我的数据如下：

3,9,2,6,4,11,4,12,2,4,Apple
4,1,4,10,3,13,3,4,1,10,Banana
2,1,2,10,4,4,4,1,4,13,Apple
2,12,4,3,1,10,1,12,4,9,Apple
1,7,3,11,3,3,4,8,3,7,Apple

最后一个值-香蕉，苹果是类索引。此属性共有9个选项。这是我在处理J48分类器的地方。

    J48 j48 = new J48();

    List<String> options = Collections.singletonList("-U");
    j48.setOptions(options.toArray(new String[0]));

    j48.buildClassifier(trainingData);
    System.out.println("J48 Classifier\n");
    System.out.println(j48);

    Evaluation j48Evaluation = processEvaluation(trainingData);
    j48Evaluation.evaluateModel(j48, testingData);

    String j48Summary = processEvaluationSummary(j48Evaluation, J48_SUMMARY);
    System.out.println(j48Summary);
    String j48Matrix = processEvaluationMatrix(j48Evaluation, J48_MATRIX_SUMMARY);
    System.out.println(j48Matrix);