---
dataset_info:
  features:
  - name: engine_no
    dtype: int64
  - name: time_in_cycles
    dtype: int64
  - name: operational_setting_1
    dtype: float64
  - name: operational_setting_2
    dtype: float64
  - name: operational_setting_3
    dtype: float64
  - name: s_1
    dtype: float64
  - name: s_2
    dtype: float64
  - name: s_3
    dtype: float64
  - name: s_4
    dtype: float64
  - name: s_5
    dtype: float64
  - name: s_6
    dtype: float64
  - name: s_7
    dtype: float64
  - name: s_8
    dtype: float64
  - name: s_9
    dtype: float64
  - name: s_10
    dtype: float64
  - name: s_11
    dtype: float64
  - name: s_12
    dtype: float64
  - name: s_13
    dtype: float64
  - name: s_14
    dtype: float64
  - name: s_15
    dtype: float64
  - name: s_16
    dtype: float64
  - name: s_17
    dtype: float64
  - name: s_18
    dtype: float64
  - name: s_19
    dtype: float64
  - name: s_20
    dtype: float64
  - name: s_21
    dtype: float64
  - name: rul
    dtype: int64
  - name: cluster
    dtype: int32
  - name: condition_ind
    dtype: float64
  splits:
  - name: train
    num_bytes: 12257052
    num_examples: 53759
  - name: test
    num_bytes: 7749948
    num_examples: 33991
  download_size: 9762754
  dataset_size: 20007000
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---


Feature Engineered version of [this](https://huggingface.co/datasets/uncledecart/fed-turbofan) dataset. Key differences:

- GBDT feature selection (selected on train data, target is rul, threshold is 0.01, lightgbm.LGBMRegressor is used)
- Sensor-wise z_score and min_max normalisation (effectively just min_max, applied z_score as in [this](https://www.researchgate.net/publication/355164116_Data-Driven_Deep_Learning-Based_Attention_Mechanism_for_Remaining_Useful_Life_Prediction_Case_Study_Application_to_Turbofan_Engine_Analysis) paper)