Task 1.1: Data Understanding: Explore the dataset with the analytical tools studied and write a concise “data understanding” report assessing data quality, the distribution of the variables and the pairwise correlations.
Task 1.2: Data Preparation: Improve the quality of your data and prepare it by extracting new features interesting for describing the player profile and his behavior derivable from matches. These indicators have to be extracted for each player. Examples of Indicators to be computed are:
- how many times did the player win during a given period
- how many matches the player played in a given period
- a ratio between the previous indicators
- percentage of aces related to the number of first serves made
- number of breakpoints numbers w.r.t. all games
- ….
Note that these examples are not mandatory. You can derive indicators that you prefer and that you consider interesting for describing the players.
It is MANDATORY that each team defines indicators and their description and when it is necessary also their mathematical formulation. The profile will be useful for the clustering analysis (i.e., the second project’s task). Once the set of indicators is computed, the team has to explore the new features for a statistical analysis (distributions, outliers, visualizations, correlations).
Subtasks of DU
- Data semantics for each feature that is not described above and the new one defined by the team
- Distribution of the variables and statistics
- Assessing data quality (missing values, outliers, duplicated records, errors)
- Variables transformations
- Pairwise correlations and eventual elimination of redundant variables
Based on the player’s profiles explore the dataset using various clustering techniques. Carefully describe your decisions for each algorithm and which are the advantages provided by the different approaches. Subtasks
- Clustering Analysis by K-means:
- Identification of the best value of k
- Characterization of the obtained clusters by using both analysis of the k centroids and comparison of the distribution of variables within the clusters and that in the whole dataset
- Evaluation of the clustering results
- Analysis by density-based clustering:
- Study of the clustering parameters
- Characterization and interpretation of the obtained clusters
- Analysis by hierarchical clustering
- Compare different clustering results got by using different merging strategies
- Show and discuss different dendrograms using the different merging strategies
- Final evaluation of the best clustering approach and comparison of the clustering obtained
- Optional (2 points): Explore the opportunity to use alternative clustering techniques in the library: https://github.com/annoviko/pyclustering/
Consider the problem of predicting for each player a label that defines if s(he) is a high ranked player or a low ranked player (binary task) by exploiting the feature related to the rank of the players.
The student need to:
- Define a player profile that enables the above player classification. For this task, you can exploit the profile created for the clustering task, by adding or removing features, depending on the results previously obtained.
- Compute the label for any customer. The extraction of the label can take advantage
of several features related to the rank, such as loser_rank, winner_rank,
loser_rank_points, winner_rank_points, etc. An example of simple label can be
derived by:
- computing the average rank per player by exploiting loser_rank, winner_rank
- select a threshold for discretizing in two categorical labels the class.
- Note that you can define in different ways the labels.
- Perform the predictive analysis comparing the performance of different models, discussing the results and discussing the possible preprocessing that you applied to the data for managing possible problems identified that can make the prediction hard. Note that the evaluation should be performed on both training and test set.
Task 4.1: Time series analysis Consider the dataset of time series CityGlobalTemperature2000-2009.csv containing for 100 cities the temperature measurements (mean and standard deviation over a month). The goal of the task is to find groups of similar cities with respect to the temperature trends.
Task 4.2: Explanation Analysis Consider the non-interpretable models used in Task 3 (eg. SVM, ensemble methods, etc) and study the global explanation with SHAP and the local explanation with LIME and SHAP. Use the evaluation metrics presented during the XAI Laboratory (for using LORE, you can download the library at the link: https://github.com/rinziv/XAI_lib_HAI-net_Tutorial and follow the instructions contained in the notebook presented during the XAI Laboratory).