ATM cash demand forecasting:

Most of the banks rely on cash-in-transit companies to replenish cash in their ATM network. The cash-in-transit companies charge their clients for counting, replenishing and borrowing money. Most of the forecasting methods these companies use are out dated, which cost a bank huge interests on the dead cash left in the ATM machine over night. Most of the times ATM machines run out of most common denominations, in this situation of cash out, the cash-in-transit companies provide taxi service at much higher rate. ATMIA has estimated these loss close to 2.7 billion dollars annually. According to ATMIA the cost of cash is between 1200 to 1500 dollars a month per ATM machine.

Banks can save 20-40% on the interest they pay to the borrower on the dead cash by using state-of-the-art forecasting solution like intelliCast.

intelliCast offered as a web service and as appliance, guarantees reduction of 20- 40% in the dead cash and reduces cash outs by over 60%.

Functional Genomics:

We have 13 years of Research and Development experience in Microarray Gene Expression analysis. Finding differential expression for a group of genes is a np-complete problem. Most of the software either work in gene space or in sample space. Very few of the software which work in both spaces either eliminate low signals or apply some statistical threshold - thus eliminating latent biological signals which are very vital in gene expression analysis.

GePIC is the only software in industry which eliminates this problem and is based on its predecessor BTSVQ. GePIC generates sample clusters based on differential expression and generates novel visualizations of highly multidimensional data into 2D.

GePIC:

> Is least sensitive to initial conditions and produces repeatable results for your experiments

> Produces novel visualizations of the very high dimensional data sets

> Generates reports in flexible formats

> Is scalable and ready to used in the cloud (uses parallel processing)

Twitter Sentiment Analysis:

The fundamental issue with opinion mining is to select the right approach for right problem to solve. The common approaches used for opinion mining are

     > parts of speech

     > n-grams using thesaurus  and

     > text mining

Each of these  approaches is used for a specific problem, for example to know the general mood of a population on some incidence or news/event, or to know sentiment for a market ,stock or company we may use the parts of speech. But we might not use parts of speech for an issue for which the thesaurus is not built yet. When we need to capture and develop thesaurus for that particular issue n-grams approach will be a good fit. For some other problems we might need to perform text mining, like clustering and classification.  The bottom-line is no single approach solves all problems.

At Predictive Analytics  we have developed a novel mechanism to handle the twitter sentiment analysis problem for all above listed approaches. This technique produces exceptionally good results (less false positives). The industry standard in twitter sentiment analysis is roughly 67% accuracy (rate of false positives over 33%) where as our technique generates over 87% correct results (false positive rate less than 13%).