1. Classification_ClaraHeine_lightgbm.txt
Algorithm: lightGBM Classifier GBDT (GradientBoostingDecisionTree)
HP values: num_leaves = 50 | max_depth = 10 
HP Optimisation: tried different settings for max_depth and num_leaves
Binary Log Loss function with all features 0.138229 (with 5fold cross validation). Accuracy: 0.9443
Execution time 22.03 seconds
The 25 variable was chosen with SHAP
Binary Log Loss function with 25 features 0.209287 (with 5fold cross validation). Accuracy: 0.9150
Execution time 4.43 seconds
Own evaluation. Well performing and fast model.

2. Classification_ClaraHeine_MLPClassifier.txt
Algorithm: MLPClassifier (NN - Adam)
HP values: solver='adam' | activation='relu' | hidden_layer_sizes=(100,) | learning_rate='constant'
HP Optimisation: tried different settings for hidden_layer_sizes
Pre-processing: Scaled with Sklearn standart scaler
Binary Log Loss function with all features 0.102407. Accuracy:  96.31
Execution time 200.54 seconds
The 25 variable was chosen with permutation importance
Binary Log Loss function with 25 features 0.144705. Accuracy: 94.54
Execution time 193.39 seconds
Own evaluation. Well performing but not fast. Better result than LightGBM-algorithm but slower. 

3. Regression_ClaraHeine_LightGBM.txt
Algorithm: LightGBM Regressor GBDT (GradientBoostingDecisionTree)
Key HP values: num_leaves = 31 | max_depth = -1 | num_iterations = 300 | learning_rate = 0.1
HP optimisation: HP was tuned using random GridSearch. Best result was in overall the default values from LightGBM it self.
MAE: 6082.55
Execution time 14.24
15 variables was chosen with in-built feature importance
MAE: 6290.69
Execution time 1.67
Own evaluation. The model performs quick and produces a result I'm satisfied with. The MAE seems a little high it self, though the energy is relatively high too.

4. Clustering_ClaraHeine_Kmeans.txt
Algorithm: sklearn KMeans
HP n_clusters=4 | init='k-means++' | max_iter=300 | n_init=10
HP optimisation: Used the Elbow method - 4 clusters
Pre-processing: Scaled with Sklearn standart scaler
8 variables was chosen by looking at feature importance in reduced dimensions with PCA
Execution time 2.96
Own evaluation: The model works and performed quickly and the result seems okay.