American-Company-Bankrupt-Prediction-Using-Neural-Network

Image Sources: Pexels.com

Business Understanding

In the realm of financial analysis and risk management, data scientists face the challenge of effectively identifying and mitigating bankruptcy risk for companies operating in dynamic economic environments. With constantly evolving market conditions and financial complexities, predicting bankruptcy accurately is crucial for investors, creditors, and stakeholders to make informed decisions and safeguard their interests 💼.

One of the primary challenges lies in navigating through vast amounts of financial data to extract meaningful insights and indicators of potential bankruptcy risk. Factors such as fluctuating market trends, industry competition, and economic uncertainties further complicate the task of identifying early warning signals and developing robust predictive models. By understanding these challenges, data scientists can leverage advanced analytical techniques and machine learning algorithms to sift through data noise and uncover hidden patterns that signify impending financial distress.

The proposed solution involves developing a predictive model for bankruptcy risk assessment using machine learning algorithms and financial indicators extracted from company financial statements. By analyzing key financial metrics such as liquidity ratios, profitability margins, and debt levels, the model aims to classify companies into distinct risk categories and provide actionable insights for stakeholders. This solution not only enhances the decision-making process for investors and creditors but also enables proactive risk management strategies to mitigate potential financial losses and maximize returns on investment 💰.

Bankruptcy Concept Video:

1. How Bankruptcy Works https://youtu.be/tpI0XWjIsqI?si=B9gLLVfw6-Hz1L34

2. Business Bankruptcy Explained: American Company Case https://youtu.be/8WB6yYsSvpE?si=lquHbIUhJlONWEm_

3. Chapter 7 Bankruptcy Explained | Step by Step https://youtu.be/MB-kdgk42h0?si=yJoITQQUOjmgMlCh

Data Understanding 📊

The dataset for the 'America Company Bankrupt Prediction' project comprises financial attributes extracted from American company financial statements. Each feature provides crucial insights into the financial health and performance of the companies, serving as essential indicators for predicting bankruptcy risk.

• nama_company: nama perusahaan
• status_label: supervised data where the data is alive or fail
• aset_sekarang (current_assets): This feature represents the total value of current assets owned by the company, including cash, accounts receivable, and inventory. Current assets are assets that are expected to be converted into cash or used up within one year.
• biaya_barang_terjual (cost_of_goods_sold): Indicates the direct costs associated with producing goods or services sold by the company during a specific period.
• penyusutan_dan_amortisasi (depreciation_and_amortization): Describes the systematic allocation of the cost of tangible and intangible assets over their useful life, representing the reduction in value of assets over time.
• laba_spbda (ebitda): Stands for Earnings Before Interest, Taxes, Depreciation, and Amortization, representing the company's operating profit before accounting for non-operating expenses.
• inventaris (inventory): Represents the value of goods and materials held by the company for production or sale.
• pendapatan_bersih (net_income): Indicates the company's total revenue minus total expenses, representing the profit or loss generated by the company during a specific period.
• total_piutang (total_receivables): Represents the total amount of money owed to the company by customers for goods or services sold on credit.
• nilai_pasar (market_value): Specifies the estimated value of the company based on its market capitalization or the market price of its outstanding shares.
• penjualan_bersih (net_sales): Represents the company's total revenue from sales of goods or services after deducting returns, discounts, and allowances.
• total_aset (total_assets): Describes the total value of all assets owned by the company, including both current and non-current assets.
• total_hutang_jangka_panjang (total_long_term_debt): Represents the total amount of debt that is due for repayment beyond the next twelve months.
• laba_sbpb (ebit): Stands for Earnings Before Interest and Taxes, representing the company's operating profit before deducting interest and taxes.
• laba_kotor (gross_profit): Indicates the company's total revenue minus the cost of goods sold, representing the profit generated from core business operations.
• total_kewajiban_lancar (total_current_liabilities): Represents the total amount of short-term financial obligations or debts that must be settled within one year.
• laba_tersisa (retained_earnings): Describes the portion of the company's net income that is retained and reinvested in the business rather than distributed as dividends to shareholders.
• total_pendapatan (total_revenue): Represents the total amount of money generated by the company from sales of goods or services.
• total_hutang (total_liabilities): Describes the total financial obligations or debts owed by the company to external parties.
• total_biaya_operasional (total_operating_expenses): Represents the total expenses incurred by the company in its normal business operations, including costs such as salaries, rent, and utilities.

Data Preparation

• Data Cleaning
• EDA
• Feature Engineering
• Feature Selection
• Encoding
• Balancing Data

Modelling

• Random Forest Classifier
• Stochastic Gradient Descent
• Gradient Boost Classifier
• Neural Network

Evaluation

• accuracy
• recall
• f1-score
• precision
• loss

Conclusion

From the modeling results before tuning, we see that some models suffer from overfitting, with low precision for failed or bankrupt companies. However, after tuning, especially on the Random Forest, Gradient Boost, and Stochastic Gradient Descent models, there is a significant improvement in bankruptcy classification, although there are still slight signs of overfitting. The application of the Neural Network also provided an improvement in model performance, with the dynamic learning rate adjustment helping to reduce overfitting and maintain stability in the validation data.

The Gradient Boost tuning model, despite having good accuracy, was found to have quite low precision and several other evaluation parameters. In contrast, the Random Forest tuning model shows an accuracy of about 75%, and also has evaluation parameters such as precision that are quite good. However, the Gradient Descent model seems to be unsuitable and unable to learn well. Nonetheless, the Neural Network Model stands out with good accuracy, showing the potential to provide more accurate bankruptcy predictions.

With more accurate information about the likelihood of company bankruptcy, they can make more informed decisions in managing their financial risks and maximizing potential profits. However, there is still a need for further research and exploration of more advanced modeling techniques to further improve model performance and ensure accuracy in bankruptcy prediction.