Competition hosted on Analyticsvidhya
Build a machine learning/deep learning approach to forecast the total energy demand on an hourly basis for the next 3 years based on past trends.
Initially, I thought a machine learning approach itself was enough to handle any time-series-based problem but this competition has proved that I was totally wrong.
In one of the previous time-series-based competitions, I used the boosting regressor model and it gave me a good leaderboard rank. I tried the same approach in this competition and it become a major blunder and had given the worst rank in the private leaderboard.
The machine learning-based approach didn't learn signals or patterns from the train data, the model learned only noise from the data.
Finally, for this problem, the boosting algorithm well performed on the train data and failed to generalize on the test data(Overfitting!).
- av-job-a-thon-november-2022-eda.ipynb
* seaborn * Pandas * Numpy * Matplotlib - av-job-a-thon-november-2022-model.ipynb
* Sklearn * Pandas * Numpy * Matplotlib * Lightgbm * shap
