This project focuses on implementing and analyzing the learning process in decision trees using Connect 4.
The dataset comprises information about houses in the Doz game filled by players (x and o). It consists of 42 features representing each house, indicating whether the player's piece is "x," "o," or neither. Based on this information, the goal is to determine the game's outcome for "x" as win, draw, or loss.
1️⃣ Implementation of the ID3 algorithm:
Implemented the ID3 algorithm using three measures, Information Gain, GINI and Log Loss to determine the best feature at each step.
2️⃣ Varying the data sharing ratio:
Considered changing the ratio between the test and training data. Explored the impact of different data sharing ratios on the decision tree model.
3️⃣ Addressing overfitting:
Examined the occurrence of overfitting and assessed its impact on the model's accuracy.
Plotted the accuracy based on tree size for both the training and test datasets. Additionally, visualized the decision trees created for the test and training data.
4️⃣ Reduced error pruning method:
Implemented the reduced error pruning method to reduce the structural complexity of the decision tree or address possible overfitting.