rainfall_prediction_by_LSTM

This is a program to classify and predict 7 classes based on 7 ranges of rainfall in An Giang, Vietnam

In this project we seperate original data based on daily period into 7 ranges to classify them:

Table of label key

Label Key	Label Name	Range of value
0	No rain	0
1	Insignificant rain	(0,0.3]
2	Light rain	(0.3,3]
3	Rain	(3,8]
4	Medium rain	(8,25]
5	Heavy rain	(25,50]
6	Very heavy rain	(50, +inf)

This project has some problems need to solve:

• Imbalanced data

  • To improve imbalanced data we need to use oversampling techniques such as SMOTE, ADASYN, SVM SMOTE,...
  • In this poject we use SVM SMOTE to oversample data

    from imblearn.oversampling import SVMSMOTE
    from collections import Counter
    import numpy as np
    
    def svm_smote(X, Y, random_state):
      #minority class count is optional but we need class count to ensure dataset is not missing any classes
      minority_class_counts = [count for _, count in Counter(np.argmax(Y, axis=1)).items() if count <= 5]
      k_neighbors_val = min(5, min(minority_class_counts) - 1) if minority_class_counts else 1
      #SVM SMOTE processing
      svmsm = SVMSMOTE(random_state=random_state, k_neighbors=k_neighbors_val)
      X_res, Y_res = svmsm.fit_resample(X, Y)
    
      return X_res, Y_res

• See my experiments click here