Project: Knowledge Representation and Probabilistic Reasoning in AI

Implements an integrated artificial intelligence framework combining First-Order Logic (FOL) inference, Bayesian Network modeling, and Naïve Bayes learning, bridging symbolic and probabilistic reasoning under the AIMA Python environment.
The project demonstrates unified representation, inference, and learning across declarative logic and uncertainty-based models.

---

Part 1 – First-Order Logic Knowledge Base and Inference

Constructed a declarative knowledge base using Horn clauses and logical entailment for reasoning over structured relational domains.
Implemented both forward and backward chaining inference strategies to perform logical deduction, unification, and variable substitution across complex rule sets.

Technical Focus

• Defined recursive Horn clauses for hierarchical relationships and transitive reasoning.
• Implemented inference using AIMA’s FolKB, fol_fc_ask, and fol_bc_ask modules.
• Compared reasoning efficiency and completeness between chaining approaches.
• Evaluated the scalability of inference across increasing rule base sizes.

---

Part 2 – Bayesian Network Modeling and Probabilistic Inference

Developed a Bayesian Network capturing dependencies between socioeconomic and environmental variables such as innovation, education, and sustainability metrics.
Performed both exact inference (enumeration, variable elimination) and approximate inference (prior sampling, likelihood weighting, Gibbs sampling) to evaluate posterior distributions and probabilistic dependencies.

Technical Focus

• Designed and parameterized conditional probability tables (CPTs) for all nodes.
• Verified conditional independence and causal consistency of the network structure.
• Quantified marginal and conditional probabilities under multiple evidence scenarios.
• Visualized probabilistic dependencies using NetworkX and Matplotlib DAG plots.

---

Part 3 – Naïve Bayes Classification and Data-Driven Reasoning

Implemented supervised probabilistic learning using Naïve Bayes classifiers on real-world datasets (Iris, Maternal Health Risk).
Extended the AIMA Naïve Bayes framework with customized data preprocessing, imputation, and posterior-distribution visualization.

Technical Focus

• Integrated data handling and imputation using Pandas and NumPy.
• Trained models via maximum-likelihood estimation with Laplace smoothing.
• Evaluated accuracy, confusion matrices, and posterior reliability across datasets.
• Produced interpretable probabilistic visualizations with Seaborn and Matplotlib.

---

Technical Overview

• Frameworks: AIMA Python, Pandas, NumPy, Matplotlib, Seaborn, NetworkX, Scikit-learn
• Core Methods: First-Order Logic inference · Bayesian Networks · Probabilistic Learning
• Conceptual Scope: Logical reasoning · Probabilistic inference · Uncertainty modeling · Explainable AI

Developed to demonstrate a complete hybrid reasoning architecture — integrating symbolic logic, probabilistic graphical modeling, and machine-learning-based classification within a unified experimental platform.